The Staff Engineer’s Playbook: Intellectual Shift to Systemic Impact  🔗 **Source:** [The Staff Engineer’s Playbook: Intellectual Shift to Systemic Impact](https://www.youtube.com/watch?v=14DjvNADo28) > [!note]+ Document Notes > tags: [staff-engineering, systems-thinking, problem-definition, goodharts-law, ground-truth, organizational-alignment, silent-mentoring, monzo] > author: Unnamed Monzo staff engineer — byline lost in capture; see Attribution > source: https://www.youtube.com/watch?v=14DjvNADo28 > reference: "The Staff Engineer's playbook: Intellectual shift to systemic impact" — conference talk > date created: 2026-08-26 18:45 > > > [!info] Generated analysis — not the speaker's words. Sections below are machine-written > > from the captured transcript. Only text inside `> [!quote]` blocks is verbatim from the speaker. > > Do not quote or attribute any other line to the named speaker. > > # 🏷️ Title: The Staff Engineer's playbook: Intellectual shift to systemic impact > > **Attribution.** Reader's `author` field reads `youtube.com`. **The speaker is not named in the transcript**, but is identifiable in outline: a **staff engineer at Monzo**, formerly a **classically trained singer and pianist** who took a forced cross-discipline module in computer science, fell into it, and completed **a PhD combining computer science and music theory**. **The byline is recoverable** — this same person published the written version as a LeadDev article, also in this batch (`01m0z4v3sqakvns54eekgysb19`), and LeadDev bylines its pieces. **Attribute to "a Monzo staff engineer" until the name is confirmed.** > > One named person appears **with explicit permission**: **Tom, Monzo's group CFO**, quoted from an internal fireside chat. > > **Register.** A conference talk, delivered as a personal narrative with a derived framework. **Not research** — the playbook comes from interviews across Monzo's IC and management tracks plus reflection on his own projects. He's clear about that: *"this isn't a checklist. It's a way of working."* > > **Why it matters here.** Two reasons. The **conviction phase** independently reproduces Lachance's ground-truth argument in a completely different domain — measurement rather than documentation. And the **validation phase is the direct counterpoint to Sanja Grbic's "less friction, less negotiation, less trying to convince"** — I flagged in that analysis that removing persuasion also removes review, and that neither speaker noticed. **This speaker treats persuasion as the actual job.** > > **Bonus finding.** This talk and its LeadDev written version are the **same author saying the same thing in two formats**, which makes them a controlled test of the compression pattern found earlier in this batch. **The written version loses a lot.** Detailed in the companion analysis. > <!-- deep-read:sections start --> ## 🧭 Table of Contents 1. Summary 2. Main Content and Notes 3. References 4. Paraphrase 5. Newsletter 6. Action Items 7. Hypotheses 8. Methodology 9. Questions 10. Future Research 11. Implications 12. Learning Styles Assessment 13. Accessibility Assessment 14. DEIB Assessment 15. Logical Validity 16. DARVO Analysis 17. Conclusion 18. Refactor Appendix 19. Template Library ## 🔗 References (in-page) - **Scott Hanselman** — cited for a specific technique: asking what acronyms mean in meetings *"not because he doesn't know what they mean, because he does"* but to create psychological safety for others. - **Tom, Monzo's group CFO** — quoted with stated permission from an internal fireside chat on mentorship and sponsorship. - **The trust triangle** — credibility, reliability, safety. Referenced as pre-existing (*"You might have seen this triangle before"*), source unnamed. - **Betamax versus VHS** — the one-line aside on better ideas losing. - **Goodhart's Law** — the phenomenon is described in detail without the term being used. - **The playbook itself** — offered as a single slide for the audience to photograph. **Not in the transcript**, which is a real loss. ## 📝 Summary **The origin story.** Six months into Monzo he gave a talk on BigQuery query optimisation — useful work, still in place. At the pub afterwards a director asked *"How did I think this presentation went?"* and then about his background. The realisation: > *"none of that, none of my journey, none of how I thought about the work had come through."* Before tech he was a **classically trained singer and pianist** who discovered he loved music theory — *"structure, pattern identification… finding themes, how things fitted together and evolved over time."* A forced cross-discipline module put him in a computer science program: *"I was the musician in the room. I'd never studied coding."* He recognised the same shapes — *"You're finding patterns. You're looking at systems, seeing how they evolve over time"* — and did a PhD combining the two. **Then the loss.** *"I identified as a creative thinker. But after my PhD, I decided to go into tech. And on that journey, somewhere along the way, I stopped being creative and I stopped using this thinking that I had learned."* **Why**, and this is the most human part of the talk: *"early on in my career in tech, I felt like a huge impostor. I was a musician sat in a room full of engineers. And I was getting advice, be more technical, be more precise, be correct, make sure everything you deliver has zero errors in it."* His verdict is careful: *"some of this advice was right… But somewhere along the journey, this advice becomes incomplete."* **The director's actual question**, which he credits as the catalyst: *"what could your background and identity as a creative thinker do to aid you in your next stage of your career?"* **The reframe:** *"A movement from reacting to deciding."* > *"Great engineers don't just react to what's in front of them. They notice patterns, contradictions where effort and outcome don't match and they step back, understand the system. They form hypotheses and they test them like a PhD, like a researcher."* **How the playbook was built.** He interviewed across the IC and management tracks — *"your principles, your directors, your VPs"* — asking *"how they think and then what do they do about how they think."* **The four moves that recurred:** 1. **They identified the real problem.** 2. **They developed conviction about the right solution.** 3. **They got people to move with them.** 4. **They made the thinking scale beyond themselves.** *"this isn't linear. You go back, you refine, you replay phases."* **Phase 1 — the research gap.** Asking people across operations, engineering, data, and product what their pain points were, one thing recurred: nobody could define how much time was spent resolving a customer problem. **Eight historical definitions.** Numbers that didn't reconcile between forecasting and dashboards. **And the signal that mattered — a Goodhart failure, described without the name:** > *"this metric was operationally important and people were optimizing around it. And this meant when the metric was wrong, the system started teaching people the wrong behaviors. For example, people were assigning work to someone else so that it would change the metric assigned to them and they would show as lower value in the system."* The diagnosis: *"everyone thought that their definition measured actual work… But in reality, all the metrics were measured via proxies. They were signals around the signal."* And: *"All of them look precise, but they were fragile. And that fragility broke trust because we weren't measuring what we thought we were."* **The mindset for this phase:** *"You're not trying to solve it. You're just trying to understand it. You want varying opinions."* The questions are *"deceptively simple"* — *"why does this matter? Why do we do it this way? What are we really measuring?"* And the **Hanselman technique** for making that safe: ask what the acronym means even when you know, so others can. > *"organizations often stop questioning inherited systems… usually visible problems aren't the real problem. For example, a performance issue might be a measurement problem. A bad operational behavior might be a broken system design. You need to find the problem beneath the problem."* **Phase 2 — conviction, and this is where it meets Lachance.** Available options: a data solution, standardising on one of the eight definitions, or improving the proxies. *"But all of that still would have inferred the truth."* > *"the principal question became why wouldn't we make this part of the system itself? Why wouldn't we have the system tell us?"* **The costs are stated honestly:** backend dependencies across system parts, product changes, coordination across distributed engineering teams, organisational friction, and **an unfixable migration limit** — *"We'd only be producing the signal from now onwards. We couldn't replay history."* The payoff: *"a metric a property of the system itself and not an interpretation layer on top of it."* **And the delivery constraint that makes it possible:** *"you don't need to just define the end path. You must define a stage plan that's going to deliver you incremental value along the way because that will mean people believe you and they come with you on this journey."* **Phase 3 — validation and alignment.** *"this metric had years of operational history. There was fear, there was secrecy. Everyone was uncomfortable on how it was calculated."* Tech, data, and product aligned quickly; **operational trust took far longer.** The problem became *"how do we move responsibly while bringing people with us as fast as possible."* **The mechanism was disclosure:** they shared how the metric worked, where it had failed before, where it was failing now, **when they made mistakes**, what they were uncertain about, and how decisions were being made. > *"transparency didn't eliminate the disagreement, but it made the disagreement discussible."* **The trust triangle** — credibility (*"Do you know what you're talking about?"*, proved through prototypes and showing working), reliability (*"do you follow through?"*, built by incremental delivery), safety (*"Do people feel comfortable engaging with you?… Do they feel heard?"*). **And the definition worth stealing:** > *"people will say people are committed to their idea, but I found unless it's on the road map, unless it has time, unless you have backing and other priorities have moved to make space for your one, you don't have commitment."* Plus: *"not be attached to your idea because the goal isn't your idea winning. It's the right solution emerging for the team."* And a realist aside: *"better ideas and history have not always won. Betamax versus VHS."* **Phase 4 — multiplication.** *"it stopped being about what I could do and started being about what the organization could do without me in the room."* Concretely: regular updates, transparent rollouts, openly discussed failures — and **a modal exposing the backend reporting to frontline staff** so the people living in the tool could verify accuracy. *"They could come to us before the problem arose."* > *"through modeling all these behaviors something interesting started happening. The questions I were asking at the beginning started spreading. I'd hear people in meetings ask the questions I had asked."* **The mechanism, and this is the talk's best idea:** > *"Your thinking spreads not through instruction but by example. Because I have found that multiplication can't really be taught. It's observed."* **Tom, Monzo's group CFO, quoted with permission:** *"the best mentor I have ever had didn't even know I was their mentee."* The speaker's term for it: **silent mentoring** — *"what you mentor becomes how the organization thinks."* **And the honest bit about ego:** *"early on being needed feels good, right?… having people rely on you and have you be a dependent in the system is a good feedback loop. You feel like you're doing your job well. But if you are so indispensable to the system, you can't step away to have greater impact."* **The close.** *"for a long time I thought creativity and technical work were separate things. But eventually I realized creativity is how you see systems. It's noticing patterns, seeing contradictions, imagining something better, and then helping other people see it, too."* ## 📖 Main Content ### 🗒️ Notes **The catalyst.** - Six months into Monzo; a technical presentation on **BigQuery query optimisation** — efficiency and cost. - At the pub afterwards, a **director** asked how he thought it went, then asked about his background. - *"what became really clear is that none of that, none of my journey, none of how I thought about the work had come through."* **Background.** - **Classically trained singer and pianist.** Went to university to study performance. - Discovered a love for **music theory**: *"I loved structure, pattern identification. I loved finding themes, how things fitted together and evolved over time."* - **A forced cross-discipline module** placed him in computer science. *"I didn't get a lot of choice… So, yep, I was the musician in the room. I'd never studied coding. I didn't know what I was doing."* - *"Very quickly, I fell in love with it. I noticed the similarities. You're finding patterns. You're looking at systems, seeing how they evolve over time."* - **PhD combining computer science and music theory.** *"I defined problems. I formed hypothesis and I tested them. I was pushing at the edges of what we understand."* - *"all of that meant I identified myself that way. I identified as a creative thinker."* **The loss, and its cause.** - *"after my PhD, I decided to go into tech. And on that journey, somewhere along the way, I stopped being creative and I stopped using this thinking that I had learned."* - **Impostor syndrome, named directly:** *"early on in my career in tech, I felt like a huge impostor. I was a musician sat in a room full of engineers."* - **The advice he absorbed:** *"be more technical, be more precise, be correct, make sure everything you deliver has zero errors in it."* - **His careful verdict:** *"some of this advice was right. You do need to be really technically strong to be a great senior engineer. But you see, somewhere along the journey, this advice becomes incomplete."* - What becomes necessary instead: *"understanding which problems matter… identify what's actually true… make tradeoffs for the organization. Help them decide what they should and shouldn't be doing. And you have to start taking on problems where the answer isn't clear yet, like a PhD."* - **The director's question:** *"what could your background do for Monzo and do for yourself? What could your background and identity as a creative thinker do to aid you in your next stage of your career?"* **The reframe.** - *"starting with what's in front of you, stepping back to understand the system and deciding what actually matters. **A movement from reacting to deciding.**"* - *"Great engineers don't just react to what's in front of them. They notice patterns, contradictions where effort and outcome don't match and they step back, understand the system. They form hypotheses and they test them like a PhD, like a researcher."* **Method.** - *"I treated it like a research problem."* - Interviewed *"leaders across the IC path, the manager path, your principles, your directors, your VPs."* *"My calendar was a mess."* - **The question asked:** *"how they think and then what do they do about how they think."* - **Four recurring moves:** identify the real problem · develop conviction about the right solution · get people to move with them · make the thinking scale beyond themselves. - *"like anything, this isn't linear. You go back, you refine, you replay phases."* - **Disclosure worth noting:** *"you're going to get used to my story slides. Yes, they're AI generated."* **Phase 1 — the research gap.** - He asked across **operations, engineering, data, and product**: *"what are your pain points, what are you struggling with, what are the team not giving you that you need right now."* - **The recurring answer:** nobody could clearly define how much time was spent resolving a customer problem. **Different definitions across the business** — later stated as **eight**. *"We would put one number into forecasting and a different one would show up on someone else's dashboard."* - **The Goodhart signal:** *"this metric was operationally important and people were optimizing around it. And this meant when the metric was wrong, the system started teaching people the wrong behaviors. For example, people were assigning work to someone else so that it would change the metric assigned to them and they would show as lower value in the system."* - **The diagnosis:** *"everyone thought that their definition measured actual work that was effort to resolve a problem. But in reality, all the metrics were measured via proxies. They were signals around the signal. But what we needed to was just measure the work."* - *"All of them look precise, but they were fragile. And that fragility broke trust because we weren't measuring what we thought we were."* - **The research gap, defined:** *"You're looking for things that don't quite make sense, where effort and outcome don't match and where the system isn't behaving the way you expect."* - **The mindset:** *"exploration. You're not trying to solve it. You're just trying to understand it. You want varying opinions. You're triangulating across the business and also across the different directions you could take."* - **The questions:** *"why does this matter? Why do we do it this way? What are we really measuring?"* - **On naive questions and the Hanselman technique:** *"simple questions can feel really naive… I love the example Scott Hanselman has given previously… where he says he'll ask what acronyms mean in meetings. And that's not because he doesn't know what they mean, because he does. It's because he wants to create that psychological safety for others to be able to ask those simple questions, too."* - *"organizations often stop questioning inherited systems. And usually visible problems aren't the real problem. For example, a performance issue might be a measurement problem. A bad operational behavior might be a broken system design. You need to find the problem beneath the problem."* - *"once you see that, you can't unsee it because then you'll be asked a much harder question. What would the right version of this look like?"* **Phase 2 — conviction.** - **Options considered:** a data solution · standardise on one of the eight historical definitions · make the proxies and heuristics more accurate. - *"But all of that still would have inferred the truth."* - **The pivot:** *"the principal question became why wouldn't we make this part of the system itself? Why wouldn't we have the system tell us? So we moved it to the back end."* - **Costs, stated plainly:** *"moving the logic into the back end meant dependencies, dependencies between all of these different parts of the system. Product changes. We also had to do even more coordination between many different teams. Our engineering teams are distributed."* - **The irreversible limit:** *"There was migration risk. We'd only be producing the signal from now onwards. We couldn't replay history."* - Plus *"a lot of organizational friction."* - **The payoff:** *"we could finally make a metric a property of the system itself and not an interpretation layer on top of it."* - **Conviction defined:** *"not just about having an idea… it's defining the principal way to fix the system. What's the right solution? Where should its source of truth be? It might not be the easiest path and you will have to balance the necessity of speed versus the perfect solution."* - **The staging requirement:** *"you don't need to just define the end path. You must define a stage plan that's going to deliver you incremental value along the way because that will mean people believe you and they come with you on this journey."* **Phase 3 — validation and alignment.** - **The emotional terrain:** *"this metric had years of operational history. There was fear, there was secrecy. Everyone was uncomfortable on how it was calculated. And at Monzo, a company that defaults to transparency, this felt so uneasy. None of us felt comfortable or safe."* - *"we did triangulate, we listened, we heard people's opinions, but at the same time we had to give the team space to lead the decision making."* - **The realist aside:** *"better ideas and history have not always won. Betamax versus VHS. Just leave that one there."* - **On messiness:** *"you're not always going to have a comfortable timeline. Your system might be causing harm. You might have operational pressure, meaning you have to move before you're going to get full agreement."* - **The asymmetry they hit:** *"tech, data, and product came together pretty quickly. However, operational trust took us a lot longer."* - **The question:** *"how do we move responsibly while bringing people with us as fast as possible."* - **What they disclosed:** *"we shared how the metric was working, where it failed in the past, where it was failing now, when we made mistakes, and what we were uncertain about, and also how we were making decisions moving forward."* - **The result:** *"transparency didn't eliminate the disagreement, but it made the disagreement discussible. And that's what allowed trust to grow."* - *"work in this phase scales because of relationships. It works because we connected it to the incentives of the teams around us. It moved because of sponsorship. We had very senior stakeholder buy-in into this complicated project."* - **The trust triangle:** **credibility** — *"Do you know what you're talking about? We proved that through prototypes, through data, through showing our working"*; **reliability** — *"do you follow through?"*, built by iterative incremental value; **safety** — *"Do people feel comfortable engaging with you? Can you collaborate with them? Do they feel heard?"* - *"it's really important to remember to not be attached to your idea because the goal isn't your idea winning. It's the right solution emerging for the team."* - *"validation isn't about being right. It's about making your thinking believable and it becomes real when people commit."* - *"You need to take your hypothesis and expose it to the organization. That means exposing your thinking, asking for challenge and iterating in the open."* - **Commitment has shape:** *"people will say people are committed to their idea, but I found unless it's on the road map, unless it has time, unless you have backing and other priorities have moved to make space for your one, you don't have commitment. You've got to turn your good idea into a shared direction for the team."* **Phase 4 — multiplication.** - *"once something works, the challenge becomes making it last. And that's where the work really changed for me personally because it stopped being about what I could do and started being about what the organization could do without me in the room."* - **Practices:** regular updates, transparent rollouts, *"openly discuss our failures and uncertainty."* - **The frontline modal — the most concrete artifact in the talk:** *"we decided to expose more of the operational system to our frontline staff that were living in the tool every day. So we made a modal so that the team could see what we were reporting in the back end. So that our frontline staff could tell us whether we had accuracy there. They could come to us before the problem arose by us forecasting the wrong staff in the end."* - *"through modeling all these behaviors something interesting started happening. The questions I were asking at the beginning started spreading. I'd hear people in meetings ask the questions I had asked."* - *"we weren't just changing a single calculation. We started changing how people thought about operational systems, measurement and trust."* - **The mechanism:** *"Your thinking spreads not through instruction but by example. Because I have found that multiplication can't really be taught. It's observed."* - **Tom, group CFO, with permission:** *"the best mentor I have ever had didn't even know I was their mentee."* - *"that's called silent mentoring because what you mentor becomes how the organization thinks."* - **On the ego cost:** *"a bit of that feels uneasy because early on being needed feels good, right? As an early engineer, having people rely on you and have you be a dependent in the system is a good feedback loop. You feel like you're doing your job well. But if you are so indispensable to the system, you can't step away to have greater impact. You can't step away to take on the bigger challenges."* - *"move from being indispensable to building something that doesn't depend on you. That's multiplication. Not scaling yourself, but scaling how the organization thinks."* - *"your impact won't just be what you deliver. It will be what continues without you."* **The close.** - *"when I started my career, I thought being good at this job was executing well on harder and harder problems. And I came to realize it's not just what you build. It's how you think."* - **The arc:** *"move from solving problems to identifying what's worth solving. Working in the system to understanding the system and then shaping the system so that they don't need you."* - **The pattern restated:** see the system clearly → define the right solution, *"the one that fixes the problem at source"* → make it believable so others choose to move → make it scale without you. - *"for a long time I thought creativity and technical work were separate things. But eventually I realized creativity is how you see systems. It's noticing patterns, seeing contradictions, imagining something better, and then helping other people see it, too."* - The full playbook is offered as **one slide to photograph** — *"I've captured the practical steps that you can take."* **Not in the transcript.** ## 📚 References - **The playbook slide** — **the most valuable missing artifact.** He explicitly says it contains *"the practical steps that you can take to implement this in your day-to-day work"*, and a transcript reader gets none of it. - **The LeadDev written version** (this batch, `01m0z4v3sqakvns54eekgysb19`) — same author, same material, **substantially thinner.** The comparison is itself a finding. - **Scott Hanselman's acronym technique** — attributed to prior presentations, not linked. **Small, concrete, immediately usable.** - **The trust triangle** (credibility, reliability, safety) — presented as known prior art with no attribution. - **Goodhart's Law** — the mechanism is described precisely and never named. - **Cross-batch pairings, and they are the reason this source matters:** - **Lachance** — *"why wouldn't we have the system tell us?"* is the ground-truth argument in a different domain. - **Sanja Grbic** — the direct counterpoint on persuasion and negotiation. - **Cunningham** — *"multiplication can't really be taught. It's observed"* supplies a mechanism for her transmission problem. - **DX** — filtered Goodhart-gamed companies *out* of their study; this speaker treats the gaming as the diagnostic signal. ## 🗣️ Paraphrase **On the idea that connects this to Lachance, which is the strongest cross-source finding here.** Faced with a metric nobody trusted, the available fixes were all refinements of inference — better heuristics, more rules, one canonical definition out of eight. He rejected all of them on one question: *"why are we trying to reconstruct the truth afterwards?"* **That is Lachance's argument about reference documentation, in a different domain.** Lachance: *"Even if you can generate a perfect first draft, you've only postponed the problem"* — reference docs must **derive** from a machine-readable source rather than be reconstructed. This speaker: a metric should be *"a property of the system itself and not an interpretation layer on top of it."* **Two independent practitioners, two domains — documentation and telemetry — one principle: derive at source, don't reconstruct after.** Neither knows about the other. And this speaker adds the honest part Lachance's version lacks: **it costs more.** Backend dependencies, cross-team coordination, organisational friction, and an irreversible limit — *"We'd only be producing the signal from now onwards. We couldn't replay history."* **On the passage that answers a question I left open elsewhere in this batch.** Sanja Grbic described AI removing coordination overhead as pure gain — *"less friction, less negotiation, less trying to convince."* I flagged that persuasion and negotiation are **also how bad ideas get filtered and how other people's context enters a design**, and that neither speaker in that episode noticed. **This speaker treats that work as the job.** *"Some of the most valuable work on that project wasn't technical."* He asks for challenge deliberately — *"exposing your thinking, asking for challenge and iterating in the open"* — and reports the result as trust rather than delay: **"transparency didn't eliminate the disagreement, but it made the disagreement discussible."** **That is the missing half, supplied independently.** The friction wasn't overhead; it was how a proposal with years of fear and secrecy attached to it became something an organisation would adopt. **Removing it would have removed the adoption.** **On "commitment has shape," which is the most immediately usable thing in the talk.** *"unless it's on the road map, unless it has time, unless you have backing and other priorities have moved to make space for your one, you don't have commitment."* **This is a checkable test for a word people use loosely**, and it belongs with the enforcement-boundaries material that has now recurred six times in this batch. **Stated agreement is not commitment; reallocated resources are.** You can verify roadmap placement, allocated time, and displaced priorities. You cannot verify enthusiasm. **Seventh instance of the same structure: the binding evidence is what changed, not what was said.** **On the Goodhart example, and why it's better than DX's treatment.** *"people were assigning work to someone else so that it would change the metric assigned to them and they would show as lower value in the system."* **DX filtered exactly this behaviour out of their longitudinal study** — companies *"incentivizing PR throughput, which is not a great idea. It can be sort of gamified."* Correct methodologically: gamed metrics are noise if you're measuring tool impact. **This speaker treats the gaming as the signal.** People routing work around a metric told him the metric was wrong before any analysis did. **Same phenomenon, opposite epistemic use** — noise to a researcher, diagnosis to a practitioner. Both are right, and holding both is more useful than either. **On silent mentoring, which supplies a mechanism Cunningham's argument needs.** *"Your thinking spreads not through instruction but by example… multiplication can't really be taught. It's observed."* Cunningham argued this week that knowledge is failing to transmit culturally because *"there's just not the time in system."* **She named the failure; this speaker names the channel.** If transmission runs through **observation rather than instruction**, then it requires proximity and shared work — **which is precisely what the squeeze removes.** **Put together they're stronger than either alone:** the transmission mechanism is modeling, modeling requires people working alongside each other, and the thing being eliminated is the working-alongside. **Neither source has both halves.** **On the impostor material, which is the most honest thing here.** *"I was a musician sat in a room full of engineers. And I was getting advice, be more technical, be more precise, be correct."* And then: *"somewhere along the way, I stopped being creative and I stopped using this thinking that I had learned."* **What makes this land is that he doesn't blame the advice.** *"Some of this advice was right… But somewhere along the journey, this advice becomes incomplete."* **The failure isn't bad guidance; it's correct guidance outliving the stage it was correct for** — which is a more useful and more generous diagnosis than the usual version. **What to hold loosely.** **One person, one company, one project.** The playbook comes from interviews he conducted and cannot verify, at an organisation whose *"default to transparency"* he credits repeatedly — **and that culture is doing a lot of work the framework doesn't account for.** The project is described only as a success; no failed attempt appears. And **the trust triangle is presented as known prior art with no attribution**, which for a talk this careful about crediting Hanselman and Tom is a small inconsistency. ## 📰 Newsletter **The best definition of "commitment" I've heard this year.** From a Monzo staff engineer's conference talk on moving from solving problems to shaping systems: > *"people will say people are committed to their idea, but I found unless it's on the road map, unless it has time, unless you have backing and other priorities have moved to make space for your one, you don't have commitment."* **Stated agreement isn't commitment. Reallocated resources are.** Roadmap placement, allocated time, displaced priorities — all checkable. Enthusiasm isn't. The talk is built around a metric nobody trusted. Nobody at the company could say how long it took to resolve a customer problem: **eight competing definitions**, forecasting and dashboards disagreeing, endless investigations. The obvious fix was a better metric. **The tell was people gaming it:** *"people were assigning work to someone else so that it would change the metric assigned to them and they would show as lower value in the system."* Classic Goodhart — and worth noting he never names it, he just spots it. His actual diagnosis: *"everyone thought that their definition measured actual work… But in reality, all the metrics were measured via proxies. **They were signals around the signal.**"* And: *"All of them look precise, but they were fragile."* Then the move I'd steal. Available fixes were all refinements of inference — better heuristics, more rules, standardise on one definition. He rejected all of them: > *"the principal question became why wouldn't we make this part of the system itself? Why wouldn't we have the system tell us?"* **Stop reconstructing the truth afterwards; make the system produce it at source.** The metric became *"a property of the system itself and not an interpretation layer on top of it."* **If that sounds familiar it should** — it's exactly what William Lachance argued about reference documentation this week: derive it from source, don't generate a snapshot that drifts. **Two practitioners, two domains, one principle**, neither aware of the other. And this one is honest about the price: cross-team dependencies, organisational friction, and an irreversible limit — *"We'd only be producing the signal from now onwards. We couldn't replay history."* **The part that surprised me** was how much of the talk is about persuasion. Tech, data, and product aligned fast; **operational trust took far longer**, because the metric had *"years of operational history. There was fear, there was secrecy."* What worked was disclosure — how it worked, where it had failed before, where it was failing now, **when they made mistakes**, what they were still unsure about: > *"transparency didn't eliminate the disagreement, but it made the disagreement discussible."* I'd flagged elsewhere this week that when people celebrate AI removing *"less negotiation, less trying to convince,"* they're forgetting that persuasion is also how bad ideas get filtered. **Here's someone treating it as the actual job.** Two more worth keeping. **Scott Hanselman's technique**, cited here: ask what an acronym means in a meeting even when you know — *"because he wants to create that psychological safety for others to be able to ask those simple questions, too."* And the ending, quoting Monzo's group CFO with permission: *"the best mentor I have ever had didn't even know I was their mentee."* His term is **silent mentoring** — *"multiplication can't really be taught. It's observed."* Which is the mechanism behind something Zoe Cunningham argued this week: knowledge is failing to transmit because there's *"just not the time in system."* **If it transmits by observation, it needs proximity — and proximity is exactly what gets squeezed out.** He's also blunt about why this is hard: *"early on being needed feels good… But if you are so indispensable to the system, you can't step away to have greater impact."* **Caveat:** one person, one company, one successful project, at an organisation whose *"default to transparency"* he credits repeatedly — and that culture is doing work the framework doesn't account for. ## ✅ Action Items **DevRel** - `#action/h1` **"Commitment has shape"** is a one-line talk moment and a genuinely useful test — roadmap, time, backing, displaced priorities. - `#action/h1` **The Hanselman acronym technique** is the cheapest psychological-safety intervention in this batch: ask the naive question so others can. - `#action/h2` **"Signals around the signal"** is a compact way to explain proxy metrics to a non-technical audience. - `#action/h2` The musician-to-engineer arc is a strong narrative for anyone arguing that non-traditional backgrounds are an asset rather than a deficit to overcome. **OSPO** - `#action/h1` **"Why are we reconstructing the truth afterwards?"** applies to any system inferring what it could emit. **Same principle as derive-from-source documentation.** - `#action/h1` **Gaming behaviour is diagnostic, not just noise.** People routing work around a metric told him it was broken before any analysis did. - `#action/h2` **"Transparency didn't eliminate the disagreement, but it made the disagreement discussible"** is the right goal for governance disclosure — not consensus, discussability. **Community** - `#action/h1` **Silent mentoring** — *"multiplication can't really be taught. It's observed"* — is directly applicable to contributor development, and pairs with Cunningham's transmission argument to give it a mechanism. - `#action/h2` **The staged-delivery requirement** (*"a stage plan that's going to deliver you incremental value along the way because that will mean people believe you"*) is how large open-source proposals earn trust. **WordPress subteams** - **Core Program Team** — `#action/h1` **The most applicable subteam.** *"Commitment has shape"* is a usable test for release-cycle planning, and the staged-value requirement is how large core proposals should be structured. - **Core** — `#action/h1` **"Find the problem beneath the problem"** with the worked example: *"a performance issue might be a measurement problem."* The 7.1 incident this week is a live case — the visible problem was a plugin fatal; the problems beneath were a core type change, an unowned triage report, and untested combinations. - **Community** — `#action/h1` Silent mentoring and the Hanselman technique are both directly usable in contributor onboarding. - **Meta** — `#action/h2` The metric-at-source argument applies to any WordPress.org telemetry that currently infers rather than records. - **Documentation** — `#action/h2` **Pairs directly with Lachance:** derive from source rather than reconstruct. Same principle, different artifact. - **Support** — `#action/h2` **The frontline modal** — exposing backend reporting to the people living in the tool so they can flag inaccuracy early — is a concrete, transferable pattern. - **Core Performance** — `#action/h3` *"All of them look precise, but they were fragile"* is a caution for any dashboard built on proxies. - **Design, Accessibility, Hosting, Plugins, Themes, Training, CLI, Playground, Polyglots, Openverse, Photos, TV, Mobile, Marketing, Core AI, Tide** — **Not applicable to this source.** **The WPOCC** - `#action/h1` **Strong fit.** The impostor-syndrome material — a musician in a room of engineers, told to be more technical and more precise — is directly relevant to a self-taught and career-changing audience. **And his diagnosis is the generous one:** the advice wasn't wrong, it *"becomes incomplete."* **FAIR** - `#action/h3` Tangential: *"a property of the system itself and not an interpretation layer on top of it"* is structurally the same argument as portability guarantees living in a contract rather than in an operator's intent. **Personal / vault** - `#action/h1` **INDEPENDENTLY CONFIRMS LACHANCE, IN A DIFFERENT DOMAIN.** *"Why wouldn't we have the system tell us?"* is the ground-truth argument applied to telemetry rather than documentation. **Two practitioners, two domains, one principle — and this one prices the cost honestly.** - `#action/h1` **SUPPLIES THE MISSING HALF OF THE SANJA GRBIC ANALYSIS.** I inferred that removing persuasion removes review; **this speaker treats persuasion as the job.** *"Transparency didn't eliminate the disagreement, but it made the disagreement discussible."* Independent confirmation of a reviewer inference. - `#action/h1` **SUPPLIES A MECHANISM FOR CUNNINGHAM'S TRANSMISSION PROBLEM.** She says knowledge stops moving; he says it moves by **observation, not instruction** — which requires proximity, which is what the squeeze removes. **Neither has both halves.** - `#action/h2` **"Commitment has shape"** — seventh enforcement-boundaries instance. **Stated agreement isn't commitment; reallocated resources are.** - `#action/h2` **The Goodhart complementarity with DX** — DX filtered gamed metrics out as noise; he reads them as diagnosis. **Both correct, and holding both is the insight.** - `#action/h3` **Recover the byline** from the LeadDev article before citing this anywhere. - `#action/h3` **The playbook slide is the missing artifact** — he says it holds the practical steps and the transcript has none of them. ## 🧪 Hypotheses 1. **The research-gap hypothesis.** The problem everyone discusses is usually not the problem worth solving; the visible symptom points at a different underlying failure. *Illustrated with one detailed case; asserted as general.* 2. **The derive-at-source hypothesis.** If you repeatedly reconstruct something after the fact, it should probably be emitted at source instead. **Independently reached by Lachance for documentation.** *Reasoned, with costs stated.* 3. **The gaming-as-diagnosis hypothesis.** People optimising around a metric in ways that damage the work is evidence the metric is wrong, available before any analysis. *A concrete observed instance.* 4. **The discussability hypothesis.** Transparency doesn't produce agreement; it makes disagreement addressable, which is what allows trust. **The most quotable and most transferable claim in the talk.** *Reported from one project.* 5. **The commitment-has-shape hypothesis.** Commitment is verifiable as roadmap placement, allocated time, backing, and displaced priorities — not as stated agreement. *Practitioner heuristic, and checkable.* 6. **The observation-transmission hypothesis.** Thinking spreads by modeling rather than instruction, and *"can't really be taught."* **Supplies the mechanism Cunningham's argument needs.** *Reported; the strong version — that it can't be taught — is unevidenced.* 7. **The indispensability-trap hypothesis.** Being needed feels like success and caps your reach. *Honest, unmeasured.* 8. **The incomplete-advice hypothesis.** Early-career guidance to be precise and technical is correct and expires. *A generous framing of a common experience.* 9. **The transparency-culture confound (reviewer).** Much of what works here may depend on Monzo's stated default to transparency rather than on the playbook. *Credited repeatedly by the speaker, unexamined as a variable.* ## 📊 Methodology **Design: a practitioner talk with a derived framework.** **No data, no measurement, no control.** The framework comes from internal interviews he conducted and reflection on projects, illustrated by a single case study. | Claim | Basis | Strength | |---|---|---| | The four moves | **Interviews across IC and management tracks**, pattern-matched | **Moderate as a framework.** Real method, unverifiable sample, no interview count. | | The metric had 8 definitions and didn't reconcile | **First-hand** | **Strong for the case.** Specific and internally consistent. | | People gamed the metric by reassigning work | **Observed** | **Strong and concrete** — the best evidence in the talk. | | Metrics were proxies, not measurement | **First-hand diagnosis** | **Strong for the case**, generalised confidently. | | Moving to source solved it | **Reported outcome** | **Moderate.** No post-hoc metrics on accuracy, trust, or adoption. | | Transparency made disagreement discussible | **Reported** | **Plausible, unmeasured**, and confounded by company culture. | | Thinking spread to other teams | **Anecdotal** — *"I'd hear people in meetings ask the questions I had asked"* | **Weak as evidence, credible as observation.** | | Commitment has shape | Practitioner heuristic | **Not a finding — a usable test.** Judge on utility. | | Creativity and engineering are the same skill | Personal synthesis | **A reframing**, not a claim about the world. | **Methodological strengths.** - **The method for building the framework is stated**: interview widely across levels, ask *"how they think and then what do they do about how they think"*, look for recurrence. **That's a real procedure**, even if unverifiable. - **Costs are named alongside benefits.** Backend dependencies, cross-team coordination, organisational friction, and the irreversible *"we couldn't replay history."* **Most talks of this kind omit the price.** - **The failure of alignment is reported, not smoothed.** Tech and product aligned quickly; **operational trust took much longer**, and he says so. - **The framework is explicitly non-linear** — *"You go back, you refine, you replay phases."* - **Sources are credited where they exist** — Hanselman by name, Tom with stated permission. - **The AI-generated slides are disclosed** unprompted. - **The ego cost is admitted** — being needed feels good, and that is the obstacle. **What is not established.** - **One person, one company, one project, reported as a success.** No failed attempt anywhere. - **The interview sample is unquantified** and unverifiable. - **No outcome measurement.** Did the new metric prove more accurate? Did forecasting improve? **The project's success is asserted.** - **Monzo's transparency culture is a large uncontrolled variable**, credited repeatedly and never isolated. - **The trust triangle is unattributed** prior art. - **The strong claim that multiplication *"can't really be taught"*** is asserted, and it is the load-bearing part of the final phase. **Bias checks.** *Survivorship* — **pronounced**: a successful project told by the person who led it. *Self-reported* — everything. *Vendor interest* — none; no product, though there is reputational benefit to a well-received talk. *False precision* — **absent**; no invented numbers, which is creditable. *Anecdote generalised* — the central weakness: one case supports a general playbook. *Newer-is-better* — n/a. **The dominant caveat is confounding**: the practices are described inside a company that already defaults to transparency and has senior sponsorship available, and neither is treated as a variable. ## ❓ Questions 1. **What is on the playbook slide?** He says it contains *"the practical steps"* and the transcript preserves none of them. **The most conspicuous gap.** 2. **Did the new metric actually prove more accurate?** The project is reported as delivered; **no outcome measurement is offered.** 3. **How many people did he interview?** *"My calendar was a mess"* is the only quantification. 4. **How much depends on Monzo's transparency culture?** He credits it repeatedly. **Would this work at a company with fear and secrecy and no default to disclosure?** 5. **What happened to the historical data?** *"We couldn't replay history"* — how did the organisation handle years of incomparable prior numbers? 6. **Is multiplication really unteachable, or just usually untaught?** The strong claim carries the final phase. 7. **Where did the trust triangle come from?** Presented as known, uncredited, in an otherwise careful talk. 8. **What did the failed version look like?** No unsuccessful attempt appears in the playbook's development. ## 🔮 Future Research - **Find the playbook slide** — it's the actionable artifact and it isn't in the transcript. - **Recover the byline** from the LeadDev version and read both together. - **Read Hanselman's original framing** of the acronym technique. - **Pair with Lachance deliberately** — two domains, one derive-at-source principle, and this one prices the cost. - **Pair with Cunningham deliberately** — she has the failure, he has the mechanism. **Neither has both.** - **Look for the negative case:** someone who ran this playbook without senior sponsorship or a transparency culture. - **Trace the trust triangle** to its origin — it is widely used and rarely credited. ## 💡 Implications **Practice.** Three usable things. **"Commitment has shape"** — check the roadmap, the allocated time, and what got displaced, not the enthusiasm. **"Why are we reconstructing the truth afterwards?"** — applies to any system inferring what it could emit. And **the Hanselman technique**, which costs nothing and creates room for everyone else's naive question. **Policy.** **Discussability rather than consensus** is the right target for organisational transparency: *"transparency didn't eliminate the disagreement, but it made the disagreement discussible."* And **gaming behaviour is governance information** — people routing work around a measure are reporting on the measure. **Security.** Indirect but real. **A metric derived at source is auditable in a way an inferred one is not**, and the same holds for provenance generally. The talk's structure — *"where should its source of truth be?"* — is the right question for any claim a system makes about itself. **Education.** **The impostor material is the most valuable teaching content here**, and its framing is unusually generous: the advice to be precise and technical *"was right"* and *"becomes incomplete."* **A stage-appropriate instruction outliving its stage** is a better diagnosis than either "bad advice" or "personal failure." **For Courtney's practice.** Four things. **It independently confirms Lachance** on derive-at-source, in a different domain and with the costs priced. **It supplies the missing half of the Sanja Grbic analysis** — persuasion as the work, not the overhead — confirming a reviewer inference from an independent source. **It supplies the mechanism Cunningham's transmission argument needs** — observation, not instruction, which requires the proximity the squeeze removes. And **"commitment has shape"** is the seventh enforcement-boundaries instance and the most checkable one yet. ## 🧠 Learning Styles Assessment - **Visual** — **Central and entirely lost.** Story slides (AI-generated, disclosed), the trust triangle, and **the full playbook on one slide he invites the audience to photograph.** A transcript reader gets the narration and none of the artifact. - **Aural** — **Native and strong.** Well-structured, personal, clearly rehearsed without being stiff. - **Verbal** — **Strong.** Several genuinely compressed formulations — *"signals around the signal"*, *"a movement from reacting to deciding"*, *"made the disagreement discussible."* - **Kinesthetic** — **Moderate to strong.** The Hanselman technique, the frontline modal, the commitment test, and the trust triangle are all immediately applicable. - **Logical** — **Strong, and the best-structured talk in this batch.** Four named phases, each with a defined mindset, a worked example, and a stated exit. The case study runs through all four rather than being reintroduced. - **Social** — **Strong.** The entire second half is about relationships, sponsorship, trust, and modeling. Credits named individuals. - **Solitary** — **Weak without the slide.** The practical steps are visual. - **Naturalistic** — Not applicable to this source. **Reading level estimate.** Roughly grade 12–13. Vocabulary is accessible and the metaphors are concrete; **the assumed context is organisational** — IC versus manager tracks, staff and principal levels, roadmaps, sponsorship, forecasting. Someone outside a structured engineering org will follow the argument and miss the furniture. ## 📊 Accessibility Assessment Assessed as a **recorded conference talk with an auto-generated transcript**. 1. **Perceivable — captions (1.2.2).** Present, with minor artifacts: *"principles"* for **principals**, *"heruristics"*, *"moving the bat logic"* for *back-end logic*, *"unsorted certain"* for *unsure/uncertain*. **No proper nouns badly mangled** — better than most in this batch. 2. **Perceivable — audio description (1.2.3, 1.2.5).** **The significant failure.** He refers to slides throughout and **explicitly invites the audience to photograph the playbook slide** — *"I've captured the practical steps that you can take."* **The talk's single most actionable artifact exists only as an image**, and is described nowhere in speech. The trust triangle is named but not drawn in words. 3. **Perceivable — info and relationships (1.3.1).** N/A for spoken delivery. 4. **Operable — timing (2.2).** Recorded, pausable. **Pass.** 5. **Understandable — language (3.1).** **Good.** Plain vocabulary, concrete metaphors, jargon mostly explained in context. 6. **Understandable — predictable (3.2).** **Strong.** Four named phases, each signposted on entry and exit. **Among the easiest talks in this batch to follow without visuals** — which partly offsets the slide problem. 7. **Robust (4.1).** Extracted cleanly. **Content-level note, and a genuinely good one.** The **Hanselman technique he cites is itself an accessibility practice** — asking what an acronym means to create room for others to ask. **Lowering the cost of admitting you don't know is an inclusion intervention**, whether or not anyone frames it that way. **Highest-value remediation:** read the playbook slide aloud, or publish it as text. Everything else in the talk survives audio; that doesn't. ## 🌍 DEIB Assessment - **The non-traditional-background material is the substantive content, and it's handled well.** A **classically trained singer and pianist** who took a forced cross-discipline module, had never coded, and describes himself as *"the musician in the room"* — then completed a PhD bridging both. **The arc is presented as an asset that was temporarily suppressed**, not as a deficit overcome. - **The impostor account is specific rather than gestural.** *"I felt like a huge impostor. I was a musician sat in a room full of engineers. And I was getting advice, be more technical, be more precise, be correct."* **And crucially he does not blame himself or the advisers** — *"some of this advice was right… somewhere along the journey, this advice becomes incomplete."* **That reframing matters for anyone in an environment signalling that their prior expertise is irrelevant**, and it names a real mechanism: correct, stage-appropriate advice that nobody updates. - **The Hanselman technique is an inclusion intervention.** Asking what an acronym means when you already know **transfers the cost of ignorance from the least powerful person in the room to the most**. That is a small structural act, and he cites it approvingly with the reason attached: *"he wants to create that psychological safety for others."* - **The frontline modal is the DEIB detail most people would skip.** They exposed backend reporting **to frontline operational staff** so the people *"living in the tool every day"* could verify accuracy and flag problems early. **That treats the least senior people in the system as the most reliable source of ground truth about it** — and it's a design decision, not a sentiment. - **Whose trust had to be earned is worth noticing.** Tech, data, and product aligned quickly; **operational trust took far longer.** The operations staff — the people whose work the metric measured and who were being mismeasured — were the hardest to convince and had the most reason to be wary. **He reports this without impatience**, and the remedy was disclosure rather than persuasion. - **Silent mentoring cuts both ways, and he doesn't examine that.** If thinking spreads by observation rather than instruction, **it spreads most to people who are in the room to observe** — which advantages those with proximity, shared background, and access. **The mechanism he celebrates has the same uneven-distribution problem** flagged in the Cunningham analysis, and it goes unaddressed. - **Inclusive language.** Neutral throughout. One named individual, credited with explicit permission. - **Cultural specificity.** Entirely within a well-resourced UK fintech with senior sponsorship available and transparency as a stated default. **The playbook assumes conditions many organisations don't have**, and that assumption isn't examined. ## ✅ Logical Validity **Strong reasoning.** 1. **The framework has a stated derivation method** — interview widely across levels, ask how people think *and* what they do about it, look for recurrence. **Not just assertion.** 2. **Costs are named alongside benefits**, including one that can't be mitigated: *"We'd only be producing the signal from now onwards. We couldn't replay history."* 3. **The gaming observation is used correctly as evidence.** Behaviour that damages the work to improve the measure is diagnostic about the measure. 4. **The proxy diagnosis is precise.** *"Signals around the signal"* — and *"All of them look precise, but they were fragile"* correctly separates apparent precision from actual validity. 5. **The alignment failure is reported rather than smoothed.** Operational trust took much longer, and he says so. 6. **He states the ego obstacle honestly** — being needed feels good, and that is the barrier to multiplication. 7. **The realist aside is genuine hedging**, not decoration: *"better ideas and history have not always won. Betamax versus VHS."* 8. **The framework is explicitly non-linear**, which pre-empts the obvious objection to any staged model. **Fallacies and weaknesses.** 1. **One case generalised to a playbook** — `#fallacy/hasty-generalization`. One project, one company, one person. Mitigated by the interview method, unverifiable. 2. **Survivorship** — `#fallacy/survivorship-bias`. Success reported by its author; no failed attempt appears anywhere. 3. **Confounding culture unexamined** — `#fallacy/suppressed-evidence`. **Monzo's default to transparency and available senior sponsorship are credited repeatedly and never treated as variables.** They may be doing more work than the playbook. 4. **No outcome measurement** — the project is described as delivered and successful; **accuracy, trust, and forecasting improvements are asserted.** 5. **Unattributed prior art** — the trust triangle, presented as known, in a talk otherwise careful about credit. 6. **An unevidenced strong claim** — *"multiplication can't really be taught. It's observed"* carries the final phase and is asserted. 7. **The spreading-thinking evidence is anecdotal** — *"I'd hear people in meetings ask the questions I had asked."* **Tone of voice.** Warm, well-prepared, and structurally disciplined — the most carefully built talk in this batch. What distinguishes it is **willingness to name discomfort**: the impostor feeling, the fear and secrecy around the metric, that *"none of us felt comfortable or safe,"* and that being indispensable *"feels good"* and is the trap. **He credits others by name and discloses his AI-generated slides unprompted.** The register is a practitioner sharing something worked out rather than a thought leader issuing a framework — helped by explicit hedges (*"this isn't a checklist"*, *"this isn't linear"*, *"the playbook does get messy in reality"*). **The one place confidence outruns evidence** is the claim that multiplication cannot be taught, which is stated flatly at the point where the argument most needs it. ## 🛡️ DARVO Analysis **Not applicable to this source.** DARVO describes a pattern in an accused party's account of alleged wrongdoing — denial, attacking the accuser, reversing victim and offender. **This is a conference talk about engineering practice.** No accusation, no accused, no contested harm. **Two adjacent observations.** **First, a blame framing was available and declined — twice.** The metric had *"years of operational history. There was fear, there was secrecy. Everyone was uncomfortable on how it was calculated"*, and people were **actively gaming it** by reassigning work. **That is a story easily told as organisational dysfunction or individual bad faith.** He tells it as a system failure instead: *"when the metric was wrong, the system started teaching people the wrong behaviors."* **The gaming is treated as a rational response to a broken measure, not as misconduct** — which is both more accurate and more generous. Similarly, on his own lost creativity: the advice to be precise and technical could be blamed. *"Some of this advice was right… this advice becomes incomplete."* **Second, and this is the seventh instance in this batch:** he places the obligation on himself. Every phase ends with something *he* must do — expose the thinking, ask for challenge, share the failures, build the thing that runs without him. **Nothing is asked of anyone else.** **Recorded as: not applicable; two declined blame framings, and the batch's seventh source to locate responsibility with itself rather than elsewhere.** ## 🎯 Conclusion **Bottom line.** The best-structured talk in this batch, and the one that connects most productively to the rest of it — supplying two things I had flagged as missing from other analyses. **The move worth taking.** Faced with a metric nobody trusted and eight competing definitions, every available fix was a refinement of inference — better heuristics, more rules, one canonical definition. He rejected all of them: > *"the principal question became why wouldn't we make this part of the system itself? Why wouldn't we have the system tell us?"* **This is Lachance's ground-truth argument, in a different domain, from someone who has never heard of him.** Lachance: reference documentation must derive from source, because a generated snapshot *"only postpones the problem."* This speaker: a metric should be *"a property of the system itself and not an interpretation layer on top of it."* **And he prices what Lachance doesn't** — cross-team dependencies, organisational friction, and an irreversible limit: *"We'd only be producing the signal from now onwards. We couldn't replay history."* **The diagnostic that got him there** is better than the fix. People were **gaming the metric** — *"assigning work to someone else so that it would change the metric assigned to them"* — and he read that as evidence about the metric rather than about the people. **DX filtered exactly this behaviour out of their study as noise. He treats it as the signal.** Both are right, and holding both is the useful position. **The half of the batch this fills in.** Sanja Grbic described AI removing *"less friction, less negotiation, less trying to convince"* as pure gain. **I flagged that persuasion is also how bad ideas get filtered, and that neither speaker noticed.** This speaker treats it as the job: *"Some of the most valuable work on that project wasn't technical."* And the result he reports is not agreement but something better: > *"transparency didn't eliminate the disagreement, but it made the disagreement discussible."* **The second half it fills in.** Cunningham argued this week that knowledge stops transmitting because *"there's just not the time in system."* **She names the failure; he names the channel:** *"Your thinking spreads not through instruction but by example… multiplication can't really be taught. It's observed."* **If transmission runs on observation, it needs proximity — and proximity is exactly what the squeeze removes.** Neither source has both halves. **The most portable single idea:** *"unless it's on the road map, unless it has time, unless you have backing and other priorities have moved to make space for your one, you don't have commitment."* **Stated agreement isn't commitment; reallocated resources are** — and unlike enthusiasm, all four are checkable. **Seventh enforcement-boundaries instance in this batch, and the most operational.** **What to hold loosely.** **One person, one company, one project, reported as a success**, with no failed attempt anywhere and no outcome measurement — the new metric's actual accuracy is asserted, not shown. **The largest uncontrolled variable is Monzo itself**: a *"default to transparency"* and available senior sponsorship, both credited repeatedly and never examined as conditions the playbook depends on. And the claim that multiplication *"can't really be taught"* is asserted flatly at the point the argument most needs it. **What earns it credit anyway:** costs named alongside benefits including the unfixable one, the alignment failure reported rather than smoothed, the ego trap admitted (*"being needed feels good"*), sources credited by name, AI-generated slides disclosed unprompted, and a blame framing declined twice — the gaming read as a system failure rather than misconduct. **Next steps.** 1. **Find the playbook slide** — he says it holds the practical steps and the transcript preserves none of them. 2. **Record the two cross-source completions**: persuasion-as-work (Sanja) and observation-as-transmission-channel (Cunningham). 3. **Recover the byline** from the LeadDev version before citing. ## 🧱 Refactor Appendix ### Literature Split Pad (H3) > [!quote] `#atom/literature/h1` > "the principal question became why wouldn't we make this part of the system itself? Why wouldn't we have the system tell us?" **Lachance's ground-truth argument, in a different domain.** Stop reconstructing the truth afterwards; emit it at source. **And this version prices the cost.** > [!quote] `#atom/literature/h1` > "people will say people are committed to their idea, but I found unless it's on the road map, unless it has time, unless you have backing and other priorities have moved to make space for your one, you don't have commitment." **The most operational idea in the talk.** Four checkable conditions replacing an unverifiable feeling. Seventh enforcement-boundaries instance. > [!quote] `#atom/literature/h1` > "transparency didn't eliminate the disagreement, but it made the disagreement discussible." **The right target for organisational disclosure** — not consensus, discussability. And the direct counterpoint to *"less negotiation, less trying to convince."* > [!quote] `#atom/literature/h1` > "Your thinking spreads not through instruction but by example. Because I have found that multiplication can't really be taught. It's observed." **Supplies the mechanism Cunningham's transmission argument needs.** If it runs on observation, it needs proximity — which is what gets squeezed out. > [!quote] `#atom/literature/h2` > "this metric was operationally important and people were optimizing around it. And this meant when the metric was wrong, the system started teaching people the wrong behaviors. For example, people were assigning work to someone else so that it would change the metric assigned to them" **Goodhart, observed and unnamed** — and used as diagnosis rather than dismissed as noise. **DX filtered this exact behaviour out of their study.** > [!quote] `#atom/literature/h2` > "all the metrics were measured via proxies. They were signals around the signal... All of them look precise, but they were fragile." The cleanest available statement of the proxy-metric problem. **Apparent precision is not validity.** > [!quote] `#atom/literature/h2` > "usually visible problems aren't the real problem. For example, a performance issue might be a measurement problem. A bad operational behavior might be a broken system design. You need to find the problem beneath the problem." The research-gap move, with worked examples. **The WP 7.1 incident this week is a live case.** > [!quote] `#atom/literature/h2` > "he'll ask what acronyms mean in meetings. And that's not because he doesn't know what they mean, because he does. It's because he wants to create that psychological safety for others to be able to ask those simple questions, too." **The cheapest inclusion intervention in this batch.** Transfers the cost of ignorance to the most powerful person in the room. > [!quote] `#atom/literature/h3` > "some of this advice was right. You do need to be really technically strong to be a great senior engineer. But you see, somewhere along the journey, this advice becomes incomplete." **A generous and accurate diagnosis:** correct stage-appropriate advice that nobody updates. Better than "bad advice" or "personal failure." > [!quote] `#atom/literature/h3` > "early on being needed feels good... But if you are so indispensable to the system, you can't step away to have greater impact." The ego trap, admitted. **The obstacle to multiplication is that the alternative is rewarding.** > [!quote] `#atom/literature/h3` > "We'd only be producing the signal from now onwards. We couldn't replay history." **The unfixable cost, stated.** Deriving at source does not retroactively fix what was inferred. ### Atomic Split Pad (H4) - `#atom/zettel/z1` **Emit the truth at source rather than reconstructing it after** — Any system repeatedly inferring something it could record has a design problem, not an accuracy problem. Applies identically to metrics and to reference documentation. **The cost is real and includes losing history.** - `#atom/zettel/z1` **Gaming a measure is evidence about the measure** — People damaging the work to improve the number are reporting a fault, before any analysis will. Treating it as misconduct discards the signal. - `#atom/zettel/z1` **Apparent precision is not validity** — A number computed to many decimals from proxies is fragile in a way that looks rigorous. The question is what it measures, not how precisely. - `#atom/zettel/z1` **Commitment is verifiable; agreement isn't** — Roadmap placement, allocated time, backing, and displaced priorities are checkable. Stated support is not commitment until something else moved to make room. - `#atom/zettel/z1` **Aim for discussability, not consensus** — Disclosure that includes failures, mistakes, and open uncertainty doesn't end disagreement; it makes disagreement addressable, which is what allows trust to form. - `#atom/zettel/z1` **Thinking transmits by observation, not instruction** — What spreads is how someone approaches ambiguity and which questions they ask. This requires proximity and shared work, which is exactly what a squeezed organisation removes. - `#atom/zettel/z1` **Stage-appropriate advice expires and nobody updates it** — Guidance that was correct early becomes incomplete later. The failure is not bad advice or personal deficiency, but an instruction outliving its stage. - `#atom/zettel/z1` **Being indispensable is a rewarding trap** — Dependence feels like competence and caps reach. The move is building what runs without you, and it costs the feedback loop that made the job feel good. - `#atom/zettel/z1` **Ask the naive question when you already know the answer** — It transfers the cost of admitting ignorance from the least powerful person in the room to the most, and it is free. ### Fleeting Queue - `#atom/fleeting/f1` **CONFIRMS LACHANCE INDEPENDENTLY, DIFFERENT DOMAIN.** Derive at source, don't reconstruct after — documentation and telemetry. **This version prices the cost.** - `#atom/fleeting/f1` **SUPPLIES THE MISSING HALF OF THE SANJA ANALYSIS.** I inferred persuasion is also review; **he treats persuasion as the job.** Independent confirmation of a reviewer inference. - `#atom/fleeting/f1` **SUPPLIES CUNNINGHAM'S MISSING MECHANISM.** She has the failure, he has the channel. **Observation requires proximity; proximity is what the squeeze removes.** - `#atom/fleeting/f1` **"Commitment has shape"** — seventh enforcement-boundaries instance and the most checkable. - `#atom/fleeting/f1` **Goodhart complementarity with DX** — noise to a researcher, diagnosis to a practitioner. **Both correct.** - `#atom/fleeting/f1` **The Hanselman acronym technique** — cheapest inclusion intervention in the batch. - `#atom/fleeting/f1` **The frontline modal** — expose backend reporting to the people living in the tool. Concrete, transferable, and a real DEIB design decision. - `#atom/fleeting/f1` **Find the playbook slide.** The actionable artifact, entirely absent from the transcript. - `#atom/fleeting/f1` **Recover the byline** from the LeadDev version before citing anywhere. - `#atom/fleeting/f1` **Monzo's transparency culture is an unexamined confound** — credited repeatedly, never isolated. Carry the caveat. <!-- deep-read:sections end --> 📄 **Full text:** [[Full Document Contents/Articles/The Staff Engineer’s Playbook Intellectual Shift to Systemic Impact.md|The Staff Engineer’s Playbook: Intellectual Shift to Systemic Impact by Monzo staff engineer - byline lost in capture]] ## Highlights - people were optimizing around it. And this meant when the metric was wrong, the system started teaching people the wrong behaviors. For example, people were assigning work to someone else so that it would change the metric assigned to them and they would show as lower value in the system. And this was the signal something was off here. You see, everyone thought that their definition measured actual work that was effort to resolve a problem. ([View Highlight](https://read.readwise.io/read/01m0z7y141a2x1syr31az0j5m9)) - Note: GAMING A MEASURE IS EVIDENCE ABOUT THE MEASURE — and note the complementarity with DX in this same batch. DX FILTERED companies that incentivise PR throughput OUT of their longitudinal study as noise ("it can be sort of gamified if you're familiar with Goodhart's Law"). Methodologically correct for a researcher. This speaker reads the same behaviour as the DIAGNOSTIC SIGNAL — people routing work around a metric told him it was broken before any analysis did. Both are right; holding both readings at once is the insight. Also: he never names Goodhart, he just spots it. THIS EXAMPLE IS ABSENT FROM HIS OWN WRITTEN VERSION. #atom/zettel/z1 - also how we were making decisions moving forward. Because transparency didn't eliminate the disagreement, but it made the disagreement discussible. And that's what allowed trust to grow because work in this phase scales because of relationships. It works because we connected it to the incentives of the teams around us. It moved because of sponsorship. We had very senior stakeholder buyin into this ([View Highlight](https://read.readwise.io/read/01m0z7xhp2p9vae5yvrf9vqm6v)) - Note: SUPPLIES THE MISSING HALF OF THE SANJA GRBIC ANALYSIS. She described AI removing "less friction, less negotiation, less trying to convince" as pure gain; I flagged that persuasion is ALSO how bad ideas get filtered and that neither speaker noticed. This speaker treats persuasion as THE JOB — "Some of the most valuable work on that project wasn't technical." And the target isn't consensus, it's DISCUSSABILITY. Note also what they disclosed: how it worked, where it failed before, where it was failing NOW, when they made mistakes, and what they were still uncertain about. Independent confirmation of a reviewer inference. #atom/zettel/z1