# AI Amplifies Existing Bottlenecks Rather Than Creating Capability AI is a multiplier on what an organization already has, not a source of new capability. DORA's framing — "AI is an amplifier of everything" — is the load-bearing claim: a small bottleneck today becomes a huge one once AI raises throughput 5–10x, because the constraint was never the step AI sped up. As one source puts it, "anything that you think is a little bit of a bottleneck now, when you apply AI to it, it's going to become 10 times that." So adoption alone produces no ROI. Gartner projects 50% of IT spend going to AI next year, yet most companies can't say what they spend or whether there's any return — and when ~20% of shipped features make things worse and ~60% don't move the needle, a 20-fold jump in raw output is not value, it's more unmeasured code. The mechanism is that value lives in the human, not the tool. The gap between adoption and impact *is* what one source calls "AI immaturity": higher-skilled developers course-correct and review AI output better, so AI "raises the ceiling" — which makes fundamentals *more* important, not less. The trap is outsourcing understanding to recover the speed: "you can outsource some thinking to AI, but you can't outsource your understanding." Skipping that produces skill atrophy now and code that "quietly raises the cost of every future change." <!-- atomicity: 1 idea ✓ --> ## Why This Matters For **[[AI & the Future of Knowledge Work MOC]]** and **[[AI Cost & Tokenomics MOC]]**, this is the unifying mechanism behind a cluster of 2026 practitioner pieces (AI ROI, AI immaturity, making productivity stick): they're all instances of *amplification, not creation*. It reframes the ROI question — don't ask "did AI make us faster," ask "what bottleneck did we just amplify 10x, and was it the one worth amplifying?" It's also the supply-side twin of two demand-side notes. [[3. Resources/Permanent Notes/Rewarding AI Usage Manufactures Usage Not Value]] shows incentives producing gamed *usage*; this shows AI producing amplified *throughput* — both are Goodhart traps where the measured number (PRs, tokens, adoption %) detaches from value. And [[3. Resources/Permanent Notes/Subscription Inference Is Sold Below Cost Creating a Tokenmaxxing Arbitrage]] is the pricing layer that subsidizes the very burn this note says isn't ROI. ## Counterpoints to Hold - **"Amplifier" can excuse any disappointing result.** If AI underdelivers, blaming "your bottlenecks were bad" is unfalsifiable — it lets the tool take credit for wins and the org take blame for losses. The claim needs a measurable before/after, not a post-hoc story. - **Some capability genuinely is new.** Translation, draft-generation, and "ask the codebase" weren't available at any skill level before; treating *everything* as mere amplification understates real new affordances. - **Ceiling-raising compounds.** If AI raises the ceiling and fundamentals are strong, the amplification is the point — the note risks reading as "AI is overhyped" when it's really "AI is leverage, and leverage cuts both ways." ## Links - [[Product & Startup Instincts MOC]] — applies bottleneck amplification to product strategy and operating-model decisions. - [[3. Resources/Permanent Notes/Instincts Are Reliable Ideas Are Not So Test Small Bets on Proven Foundations]] — applies-in: small tests reveal whether increased AI throughput improved the real outcome or only amplified the existing constraint. - [[AI's Leverage Comes From the Data Foundation Not the Model]] — applies-in: a weak data foundation is the bottleneck AI amplifies into faster, more confident noise. - [[AI Erodes the WordPress Plugin Value Props Pushing Monetization Toward SaaS and Services]] — applies-in: AI magnifies the plugin market's existing uncertainty about whether support, updates, or integration depth is the value customers buy. - [[3. Resources/Permanent Notes/The Bottleneck Moved From Code Generation to Review and Judgment]] — the specific bottleneck AI amplifies in engineering: generation got cheap, so review/judgment is now the binding constraint. - [[3. Resources/Permanent Notes/Rewarding AI Usage Manufactures Usage Not Value]] — the incentive twin; same Goodhart detachment of measured number from value. - [[3. Resources/Permanent Notes/AI Compresses the Knowledge-Worker Pipeline From Both Ends]] — where the amplified-skill premium lands in the labor market. - [[3. Resources/Permanent Notes/Subscription Inference Is Sold Below Cost Creating a Tokenmaxxing Arbitrage]] — the pricing subsidy financing the amplified burn. - [[3. Resources/Permanent Notes/A Capacity Fix Sized to One Volume Driver Expires When the Driver Changes]] — the failure this amplification produces downstream: capacity sized for the old growth driver meets the new one and fails without warning. - [[3. Resources/Permanent Notes/Reclassify What Blocks Before Optimizing What Blocks]] — the response that actually relieves an amplified bottleneck: remove work from it rather than speed it up. ## Open Questions - What's the cheapest instrument that tells you *which* bottleneck you amplified before you've spent the budget — is there a leading indicator, or only the lagging ROI number? - If value lives in developer skill, does heavy AI use erode the very skill that makes AI valuable — and on what timescale does that loop bite? ## Source - [Reader: How do you know your AI investment is worth it?](https://read.readwise.io/read/01kv3seec8jbfrnfy6jk0j7q47) — Gartner 50% IT spend; 80% of features don't move the needle; DORA "amplifier"; bottleneck-becomes-10x - [Reader: The high cost of AI immaturity](https://read.readwise.io/read/01kv3sv8d5d16997b18yz6a67n) — adoption≠impact; value in the developer's skill; "more engineering discipline, not weaker" - [Reader: Make AI productivity gains stick](https://read.readwise.io/read/01kv3t2esb3g9ncyvbt9bvhn7z) — AI raises the ceiling; can't outsource understanding; skill atrophy - See also: [[AI & the Future of Knowledge Work MOC]] · [[AI Cost & Tokenomics MOC]]