With AIs Wide Open by Chris Heilmann - CourtneyR.dev's 2nd Brain

With AIs Wide Open by Chris Heilmann 🔗 **Source:** [With AIs Wide Open by Chris Heilmann](https://www.youtube.com/watch?v=l9YV5z_zYQE) > [!note]+ Document Notes > tags: [AI, open-source, web-development, developer-tools, tech-career] > author: Chris Heilmann > source: All Things Open > reference: https://www.youtube.com/watch?v=l9YV5z_zYQE > date created: 2025-01-11 22:30 > > # 🏷️ Title: With AIs Wide Open >  ## 🧰 Action Inbox - [H1] Resist paying subscription fees for AI coding tools — 100k/year hidden costs #tech-debt - [H2] Use AI for prototypes and throwaway code only — not production systems #development-practices - [H3] Demand open alternatives to proprietary AI platforms — access for all developers #open-source ## 🧭 Table of Contents - [📝 Summary](#-summary) - [📖 Main Content](#-main-content) - [🗒️ Notes](#️-notes) - [📚 References](#-references) - [🗣️ Paraphrase](#️-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action Items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future Research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning Styles Assessment](#-learning-styles-assessment) - [📊 Accessibility Assessment](#-accessibility-assessment) - [🌍 DEIB Assessment](#-deib-assessment) - [✅ Logical Validity](#-logical-validity) - [🛡️ DARVO Analysis](#-darvo-analysis) - [🎯 Conclusion](#-conclusion) ### 🔗 References - [With AIs Wide Open - All Things Open](https://www.youtube.com/watch?v=l9YV5z_zYQE) - [We Are Developers](https://wearedevelopers.com/) ## 📝 Summary - Chris Heilmann shares 30 years of web development experience and concerns about AI's impact on open development - AI coding tools create hidden subscription costs (~$100k/year per team) and technical debt - Current AI hype contradicts open source principles: restricted access, no longevity, opaque procedures - Junior developers risk losing learning opportunities when AI does everything for them - Need community resistance to protect open web from corporate AI takeover ## 📖 Main Content Chris Heilmann, a veteran developer with 30 years of experience across Yahoo, Mozilla, and Microsoft, argues that the current AI development trend threatens core open source principles. Unlike previous technology hypes that never materialized, AI presents a genuine challenge to accessible, transparent development practices. The presentation highlights several concerning trends: AI coding tools essentially function as expensive subscriptions in disguise, with small development teams facing $100,000 annual costs. These tools generate code that works initially but creates substantial technical debt, as LLMs excel at creating new code but struggle with maintaining existing systems. Heilmann contrasts his definition of "open" - access for all regardless of location or wealth, documented functionality, public code review, and longevity - with current AI practices that restrict access by geography, hide functionality behind opaque agents, and encourage throwaway development. The talk draws from Stack Overflow developer surveys showing declining trust in AI coding accuracy (from 42% in 2023 to 33% in 2025) despite increased usage, often mandated by management rather than chosen by developers. ## 🗒️ Notes #### ⏱️ [00:00](https://www.youtube.com/watch?v=l9YV5z_zYQE&t=0) — [H1] Career Journey Through Tech Changes - Key idea: 30 years of surviving technology hypes by focusing on open standards - Evidence: "I survived so many changes to the world that are basically the new thing" #### ⏱️ [02:15](https://www.youtube.com/watch?v=l9YV5z_zYQE&t=135) — [H2] Early Web Development (1994-2000) - Key idea: Started as radio newscaster, self-taught HTML and Perl in 1995 - Evidence: "He asked me what the telephone number is where he reaches his website" #### ⏱️ [04:30](https://www.youtube.com/watch?v=l9YV5z_zYQE&t=270) — [H2] Dot-com Crash Experience - Key idea: Witnessed first major tech bubble burst at eToys.com - Evidence: "Y2K happened... German part of eToys went bankrupt" #### ⏱️ [07:45](https://www.youtube.com/watch?v=l9YV5z_zYQE&t=465) — [H2] Enterprise and Government Work - Key idea: Moved to stable government contracts after dot-com crash - Evidence: "You don't have to deliver anything for 5 years because everything is so much paperwork" #### ⏱️ [10:20](https://www.youtube.com/watch?v=l9YV5z_zYQE&t=620) — [H2] Yahoo Open Source Era - Key idea: Led YUI JavaScript library and Yahoo Developer Network - Evidence: "We did Yahoo pipes or YQL... a lot of open stuff" #### ⏱️ [12:10](https://www.youtube.com/watch?v=l9YV5z_zYQE&t=730) — [H2] Mozilla Firefox OS Failure - Key idea: HTML5 phone failed due to missing WhatsApp support - Evidence: "WhatsApp didn't want to be on it... in South America, we couldn't sell the thing" #### ⏱️ [14:35](https://www.youtube.com/watch?v=l9YV5z_zYQE&t=875) — [H2] Microsoft and Edge Development - Key idea: Worked to kill Internet Explorer, built Edge developer tools - Evidence: "I want to kill Internet Explorer... we brought Edge" #### ⏱️ [16:50](https://www.youtube.com/watch?v=l9YV5z_zYQE&t=1010) — [H1] Definition of Open Development - Key idea: Open means access for all, documented APIs, public code, longevity - Evidence: "Access for all independent of location or wealth" #### ⏱️ [18:45](https://www.youtube.com/watch?v=l9YV5z_zYQE&t=1125) — [H1] AI as Anti-Open Development - Key idea: Current AI hype contradicts all principles of open development - Evidence: "The current hype around AI is almost the anti-thesis of almost all of that" #### ⏱️ [21:30](https://www.youtube.com/watch?v=l9YV5z_zYQE&t=1290) — [H1] Hidden AI Development Costs - Key idea: AI coding tools cost approximately $100k per small team annually - Evidence: "On average, this means 100k for a development team a year" #### ⏱️ [24:15](https://www.youtube.com/watch?v=l9YV5z_zYQE&t=1455) — [H1] Declining AI Trust Statistics - Key idea: Developer trust in AI accuracy dropped from 43% to 33% - Evidence: "33% of developers trust AI accuracy in 2025, 43% did that in 2024" #### ⏱️ [27:00](https://www.youtube.com/watch?v=l9YV5z_zYQE&t=1620) — [H2] AI Code Debugging Problems - Key idea: 45% report AI debugging takes more time than expected - Evidence: "Debugging AI generated code takes more time than expected" #### ⏱️ [30:20](https://www.youtube.com/watch?v=l9YV5z_zYQE&t=1820) — [H1] VIP Coding Creates Legacy Debt - Key idea: Generated code becomes unmaintainable technical debt - Evidence: "When you vip code, you're incurring tech debt as fast as the LLM can spit it out" #### ⏱️ [35:45](https://www.youtube.com/watch?v=l9YV5z_zYQE&t=2145) — [H2] Practical AI Use Case - Content Analysis - Key idea: Uses AI for video content search and moment extraction - Evidence: "Video button... moments button... playlist button" #### ⏱️ [42:30](https://www.youtube.com/watch?v=l9YV5z_zYQE&t=2550) — [H1] Junior Developer Learning Crisis - Key idea: New developers can't learn from mistakes if AI does everything - Evidence: "How can we say that junior developers learn if they're encouraged to let magical AI do everything?" ## 📚 References - [Stack Overflow Developer Survey](https://stackoverflow.com/) - [We Are Developers Congress](https://wearedevelopers.com/world-congress/) - [Video Tap AI Tool](https://videotap.com/) ## 🗣️ Paraphrase - A 30-year web development veteran argues that current AI tools threaten the fundamental principles of open source development by creating expensive dependencies and restricting access - AI coding assistance generates significant technical debt because it excels at creating new code but fails at maintaining existing systems, leading to constantly rewritten rather than improved codebases - The speaker advocates for using AI only for disposable prototypes while preserving open, documented, and community-controlled development practices for production systems ## 📰 Newsletter The WordPress community should pay attention to Chris Heilmann's warnings about AI development tools. As we build the future of WordPress, we must resist the temptation of proprietary AI coding assistants that create hidden subscription costs and technical debt. His experience shows that truly transformative technologies survive by embracing open principles: accessible APIs, documented functionality, and community ownership. WordPress has always thrived on these values - from our GPL license to our REST API to our plugin ecosystem. Consider using AI for rapid prototyping and content analysis (like Heilmann's video search tool), but maintain human expertise for core development. The WordPress project's strength lies in our transparent development process and collaborative community, not in outsourcing our craft to black-box systems. Resources: [WordPress Coding Standards](https://developer.wordpress.org/coding-standards/), [Contributing to Core](https://make.wordpress.org/core/handbook/), [Plugin Development](https://developer.wordpress.org/plugins/) ## ✅ Action Items #### DevRel - Developer advocacy: Document hidden costs of AI tools, promote open alternatives - Education: Create workshops on sustainable AI usage vs dependency - Developer experience: Maintain human-readable documentation standards - Developer marketing: Counter AI hype with practical open source benefits - Community management: Foster discussions on AI ethics and development practices #### OSPO - Policy & Governance: Establish guidelines for AI tool evaluation and procurement - License Compliance: Review AI-generated code for licensing implications - Security Management: Assess risks of AI-generated technical debt - Software Inventory: Track AI dependencies and subscription costs - Developer Support: Provide training on judicious AI usage - Community Engagement: Fund open AI research and tooling projects - Strategic Alignment: Measure productivity claims against technical debt costs #### Community - Maintainers: Resist pressure to adopt expensive AI tools without clear ROI - Contributors: Preserve learning opportunities for junior developers - Sponsors: Support open AI research rather than proprietary platforms - Ecosystem: Coordinate resistance to AI vendor lock-in #### WordPress - Core: Maintain human review standards regardless of AI assistance - Core Performance: Monitor performance impact of AI-generated code - Core AI: Focus on open, transparent AI implementations - Community: Educate about sustainable AI practices - Training: Develop curriculum balancing AI tools with fundamental skills - Documentation: Ensure AI-assisted documentation remains accurate and maintainable ## 🧪 Hypotheses - AI coding tools will consolidate into expensive subscription oligopolies similar to enterprise software - Developer productivity gains from AI are offset by increased technical debt maintenance costs - Organizations mandating AI usage will see short-term velocity gains followed by long-term maintenance burden - Open source AI alternatives will emerge to challenge proprietary coding assistants ### 📊 Methodology - Survey development teams on actual AI tool costs vs productivity benefits - Analyze technical debt accumulation in AI-assisted vs traditional development projects - Track junior developer skill acquisition with and without AI assistance - Compare maintenance costs of AI-generated vs human-written code over time ### ❓ Questions - How can open source communities compete with well-funded proprietary AI platforms? - What metrics accurately measure developer productivity beyond lines of code generated? - How do we preserve learning pathways for new developers in an AI-assisted world? ### 🔮 Future Research - Long-term studies of technical debt in AI-generated codebases - Economic analysis of AI subscription costs vs development team budgets - Educational research on skill development with AI assistance ### 💡 Implications - Practice: Evaluate AI tools for hidden costs and vendor lock-in risks - Policy: Establish procurement guidelines for AI development tools - Security: Review AI-generated code with same rigor as human contributions - Education: Maintain fundamental programming education alongside AI literacy - Funding: Support open AI research to prevent proprietary consolidation ### 🧠 Learning Styles Assessment Reading level estimate: Flesch-Kincaid Grade Level 10 #### 👀 Visual learners - Present: Career timeline, cost breakdowns, survey statistics - Improve: Add flowcharts of AI vs traditional development processes #### 🔊 Aural learners - Present: Full video presentation with clear narration - Improve: Podcast version with chapter markers #### 🗣️ Verbal learners - Present: Rich storytelling, analogies, clear definitions - Improve: Glossary of technical terms and industry jargon #### 🧩 Kinesthetic learners - Present: Practical examples of AI tools and workflows - Improve: Hands-on exercises comparing AI vs manual coding #### 🧠 Logical learners - Present: Statistical evidence, cost analyses, timeline progression - Improve: Decision trees for AI tool evaluation #### 🧑‍🤝‍🧑 Social learners - Present: Community concerns, collaborative development principles - Improve: Discussion forums, peer review opportunities #### 🧘‍♀️ Solitary learners - Present: Individual reflection on career experiences - Improve: Self-assessment tools for AI dependency #### 🌿 Naturalistic learners - Present: Ecosystem metaphors for development communities - Improve: Environmental impact of AI energy consumption ### 📊 Accessibility Assessment #### Screen reader semantics - Present: Clear heading structure, descriptive content - Improve: Alt text for any visual elements in presentation #### Keyboard navigation - Present: Video controls accessible via keyboard - Improve: Transcript with timestamp navigation #### Contrast and readability - Present: High contrast presentation slides - Improve: Ensure all text meets 4.5:1 contrast ratio #### Text resizing and reflow - Present: Responsive video player - Improve: Scalable transcript formatting #### Media and captions - Present: Full transcript provided - Improve: Professional captions with speaker identification #### Forms and interactions - Present: Standard video controls - Improve: Custom controls with accessibility labels ### 🌍 DEIB Assessment #### Cultural breadth - Present: Global perspective on technology access restrictions - Improve: Include perspectives from underrepresented regions #### Inclusive language - Present: Gender-neutral technical terms - Improve: Avoid ableist metaphors in technology descriptions #### Neurodiversity support - Present: Clear section breaks, consistent structure - Improve: Content warnings for discussions of job loss/industry disruption #### Representation - Present: Individual perspective from industry veteran - Improve: Include voices from diverse career stages and backgrounds #### Psychological safety - Present: Respectful critique of industry practices - Improve: Balanced perspective on AI benefits alongside concerns ### ✅ Logical Validity #### Strong reasoning - Deductive strength: Historical pattern analysis supports AI hype skepticism - Inductive strength: Personal experience across multiple technology cycles - Causal reasoning: Links AI tool costs to subscription business models - Analogical strength: Compares AI hype to previous failed technology promises #### Fallacies - Appeal to experience: Relies heavily on personal career history, may not generalize - Slippery slope: Suggests AI tools inevitably lead to vendor lock-in without considering alternatives #### Tone of voice - Overall tone: Skeptical but pragmatic, grounded in practical experience with specific examples ### 🛡️ DARVO Analysis - Deny: No denial patterns observed - Attack: Critiques AI industry practices, not individual accusers - Reverse victim/offender --- ## Full Transcript [00:05] So um I just want to see how we as an [00:08] open community and the people who've [00:09] been contributing it for such a long [00:11] time can now work with this new threat [00:14] of like uh of AI taking over everything [00:17] and how can we actually deal with that [00:19] on our own basis as well. Like [gasps] [00:22] in my whole career for like 30 years, [00:24] I've always been like, "Oh, you're [00:25] falling behind because you're not [00:26] embracing this new technology or that [00:28] new thing." And it never it never came [00:31] to pass. It was always a thing like [00:33] working on standards, working on open [00:34] things, and throwing things out there. I [00:37] survived so many changes to the world [00:39] that are basically the new thing. It's [00:42] it's a very interesting thing to not get [00:45] riled up about. So let me start uh by [00:48] just showing off a bit or just I was [00:50] just writing this down and confused [00:51] myself with it as well like how how I [00:54] got into this and how I what I've done [00:56] over the last 30 years in this market. I [01:00] started as a radio newscaster in 1994 [01:04] and uh I hated it after a while because [01:06] I loved being able to give people the [01:09] news and doing research and uh reporting [01:12] on things. But then I realized like [01:14] after a while people pay for news to be [01:17] broadcast that you have to say although [01:19] it's boring and you're not allowed to [01:21] say naughty words on on air and you're [01:23] not allowed to do the things that you [01:24] want to do. So I actually taught myself [01:26] HTML and Pearl and started web [01:29] development. 1995 I built my first [01:31] website for a solicitor and he basically [01:35] then like paid money for it and then he [01:37] asked me what the telephone number is [01:38] where he reaches his website. So he [01:40] didn't quite know what the internet was, [01:41] but he wanted to spend money on it. So [01:43] that was a good thing for me. And then I [01:45] was one of the few people in Germany [01:46] back then that wrote HTML and knew how [01:49] browsers break and build things. So I [01:51] get I started working at an agency and I [01:54] did things like bank websites, little [01:56] e-commercy thingies, like restaurant [01:59] websites, things like that. And then I [02:01] got into e-commerce like these out of [02:04] the box e-commerce systems like then [02:07] that basically you paid a lot of money [02:08] for and then you can build a store on [02:10] top. This is completely different [02:12] nowadays with all the online stores [02:13] being easy to do. But back then you had [02:15] to buy a software package, install the [02:17] thing, install the server and then [02:19] charge your customers to do e-commerce [02:21] on the thing. [02:23] Then etoys.com uh hired me and said like [02:26] Chris you're going to be our main [02:27] developer in Munich in Germany and [02:29] you're going to be our HTML monkey. That [02:31] was the job title actually. And uh then [02:34] then Y2K happened and they're like Chris [02:36] you got to come to America and make sure [02:38] that our website doesn't die because [02:40] things falling from the sky and things. [02:42] U nothing happened but I stayed for 7 [02:45] months in a hotel in LA and during that [02:48] time the German part of ETOS went [02:50] bankrupt. So to stay with the company I [02:52] had to move to England and that's what I [02:54] did. So I went to England and then two [02:56] months later it went bankrupt there as [02:57] well. And then I realized okay [02:59] something's going on here and that was [03:00] the first.com crash when the whole thing [03:02] basically fell on its and its ars about [03:05] not knowing like oh we have to sell [03:06] things as well to make money. That's [03:08] weird. [snorts] [03:09] And so I basically ducked down and went [03:12] work on government websites and longer [03:14] clients larger clients. Who here works [03:15] on government stuff? [03:18] It's great isn't it? you don't have to [03:19] deliver anything for 5 years because [03:21] everything is so much paperwork that you [03:23] actually cannot do it as well. It was it [03:25] was really wonderful like doing the the [03:27] the the uh tourism portal for England uh [03:31] of different different councils of [03:33] London doing these things. It was [03:35] wonderful and larger clients like [03:37] McDonald's, HP Europe, these kind of [03:40] things. The other day I found I still [03:42] have access to one of the servers after [03:43] like 20 years because they didn't change [03:45] the password. Um okay. [03:48] And then I went into enterprise content [03:49] management system things and that was [03:51] basically for the local government in [03:53] England and that was done in India [03:55] because it's cheaper there it seems. And [03:58] that was like the weirdest moment when [04:00] you pay £40,000 license for an [04:03] enterprise content management system and [04:05] then you have to send each of your [04:06] engineers to a £3,000 course to be [04:08] allowed to write code for it. It's like [04:11] printing money. It's like was really [04:13] bizarre. And then you sit in India and [04:15] people outside sit in the rain and [04:16] basically there's things falling apart [04:18] and you press a button and install a [04:20] $50,000 license of something and you're [04:22] like that's weird. That's I I don't want [04:24] to do that ever again. So I went to [04:26] Yahoo um Yahoo basically in the UK and [04:29] then also in America a few times and [04:31] that's when I started uh pushing for [04:33] open things. I did the YUI uh JavaScript [04:36] library back then and we did Yahoo pipes [04:38] or YQL if you remember that one. We did [04:41] like a lot of open stuff and I ran the [04:43] Yahoo developer network, the blog I I [04:45] wrote that and gave talks about about [04:47] everywhere in the world about Yahoo and [04:49] what we do with search and how we're [04:51] changing search from a portal to a [04:54] search engine. And then I went to [04:55] Mozilla and that was even more open, but [04:58] it was also the weird side of open [05:00] because it's academic wonderland. You [05:02] can do whatever you want, but you also [05:03] have to promote it. Like nobody helps [05:05] you to actually get the market to [05:07] embrace the things that you need to do. [05:09] And sometimes it's just like, oh, [05:10] because it's open, people will love it. [05:12] Firefox OS was the big thing I worked on [05:14] in the end, which was an HTML 5 phone. [05:18] Everything was just web standards on the [05:20] phone. And it it it utterly and [05:22] amazingly took off, failing to make any [05:24] impact in the market at all. Uh the [05:27] biggest problem was that that uh that [05:29] basically WhatsApp didn't want to be on [05:30] it. And that meant in South America, we [05:32] couldn't sell the thing. Nobody wanted [05:33] it. people they order a a Facebook phone [05:36] or a [05:38] a a chat phone rather than a phone. They [05:40] don't even know what it is. And then I [05:42] went to Microsoft which was hilarious [05:44] because Microsoft was my big hate thing [05:46] in the beginning. Uh but during my job [05:49] interview at Microsoft they asked me [05:50] what I want to do here and I said I want [05:52] to kill Internet Explorer [05:54] and uh cuz after 25 years yeah and we [05:59] brought Edge and yeah we did that we [06:01] made it then then Edge went to Chromium [06:03] and I was the product manager of the [06:04] developer tools inside Chromium that's [06:06] also shared with Chrome. So I worked on [06:08] that thing and it was actually really [06:10] exciting until I ended up on the wrong [06:11] spreadsheet and I got laid off with 11 [06:13] 11,000 people. So, okay. Now, I do I'm [06:17] at We Are Developers. We are Developers [06:19] is a community. We actually I write a [06:21] newsletter that I have to write later. [06:24] 150,000 subscribers. So far, I think [06:25] 200,000 almost. Uh we have a podcast [06:28] that I record all the time. We have a [06:30] live show that's happening every [06:31] Wednesday lunchtime in Berlin where I [06:33] live. So, tomorrow morning at 6:00 in [06:35] the hotel room, I'm going to do that one [06:38] probably in like a bathrobe and with the [06:40] mini bar strewn around me or something. [06:41] I I don't know what's going to be [06:42] happening. And we have a magazine that [06:44] we actually write articles now as well, [06:46] much like the YDN library, YDN blog back [06:49] then or the Mozilla blog that I wrote. [06:51] And we have this congress uh which is me [06:54] in a bathrobe on stage. Uh and that's [06:57] 15,000 people and 600 speakers, eight [07:00] tracks, and I'm the content curator of [07:02] the whole thing. So I go through like [07:05] 2,400 2,600 call for paper submissions. [07:08] And if you use chat GPT to write it, I [07:11] will find out where you live [07:13] because you're wasting everybody's time. [07:16] And that whole career and that whole [07:17] thing made me think what is open to me [07:19] and what does open mean? And uh to me [07:22] open means several things. First of all, [07:24] access for all independent of location [07:27] or wealth. APIs should be free to try [07:30] them out. SDKs should be free to get [07:33] started. That's something that I always [07:34] found amazing. And the biggest change [07:36] that I find in the AI world that we're [07:38] in right now, as a European, I'm treated [07:40] as a secondass citizen. I always get [07:42] access to things much much later because [07:44] we care about things like privacy and [07:46] security and that's just h we don't want [07:48] that, you know. And I got a VPN so I can [07:51] always pretend to access to things [07:52] differently. But it just fascinates me [07:54] how that became a thing all of a sudden. [07:56] That where you live is not getting you [07:58] access to the coolest, newest [08:00] technology. And that's a huge step back [08:02] and it's terrible. I think access to [08:04] functionality without docu with [08:06] documentation should be done as well. [08:07] Not like magic like just call our agent [08:10] and he will do stuff for you and you're [08:11] like I don't want to have a robot do [08:13] things to me that I don't know about [08:14] like that's not a good idea. Access to [08:16] code inbuilt quality control. What I [08:19] always told people writing open source [08:21] software is great because you cannot [08:24] hide bad code. Everything is in public. [08:27] Everybody can take a look at it. [08:28] Everybody has an opinion about it and [08:30] everybody will actually fix or extend [08:32] the things that you need to do. That's [08:35] why also when you do uh when you look [08:36] for things on Stack Overflow or ask for [08:39] things, don't ask how something works. [08:41] Just write it wrong and let people tell [08:43] you that you're wrong and give you the [08:44] right answer. It's much much faster than [08:46] trying to get explanations out there. [08:49] Longevity. People can build on my work. [08:52] I gave out everything I did for free. My [08:54] books is creative comments. my my [08:56] bloggers creative comments as well all [08:58] the software as like MIT or Apache [09:00] license because if I get run over by a [09:02] bus the thing lives on you can do [09:05] something with it that I hadn't thought [09:06] of and I I keep being uh uh surprised [09:10] and excited about people doing stuff [09:12] with my things that I hadn't thought of [09:13] and it's wonderful that these things [09:14] exist and building things that work. I [09:18] always wanted to build open source [09:19] things that work because other people [09:21] can fix them for me and find the [09:22] problems rather than just building the [09:24] one thing that gives you the happy path [09:27] and that's and and then like the MVP and [09:29] then let's make it a live thing and make [09:30] money from it. Following agreed [09:32] procedures and ensuring quality is [09:34] another thing that I love that version [09:36] controlling became so easy since git and [09:39] github as well. I remember before that [09:41] we had like oh god mercurial svn CVS [09:45] team c team like about 50,000 of those [09:48] and they're all rubbish and then the [09:50] other day we found a we found a bug in [09:52] git that's hilarious if you haven't seen [09:54] that one it's a colleague of mine from [09:56] Microsoft they had a repo that was ended [09:59] up 26 gigabyte of uh of data and they [10:03] couldn't get outside contributions [10:04] anymore because it took people too long [10:05] to fork it and actually download it and [10:08] it turned out that git [10:10] only gets the last 12 characters of a [10:12] file name to see if it's a unique file. [10:16] So change log.m MD was seen as the as as [10:20] one file and kept attaching to each [10:22] other and became bigger and bigger and [10:24] bigger. So the more folder you had, the [10:26] more the more change log you had, the [10:27] bigger the project got. So they fixed [10:29] the the bug in git and then all of a [10:31] sudden this project is now 500 megabyte. [10:35] So, it's I don't know what got him to [10:38] think 12 characters are enough. Was he [10:39] on a DOS computer or I don't know, but [10:41] it was really bizarre that these things [10:43] out there. And the current hype around [10:45] AI is almost the anti-thesis of almost [10:47] all of that. Like it's not access for [10:49] all. You pay for everything. It's not [10:51] for longevity. Just build the thing and [10:52] throw it away immediately. Again, [10:54] there's no procedures. Like just go wild [10:56] and do stuff. And it feels weird to me. [10:59] And AI AED development is of course the [11:01] big thing. I worked with the uh with the [11:03] co-pilot people in GitHub when it [11:05] started. It was really exciting about [11:06] because I worked on VS Code back then [11:08] and to me every AI IDE that I see there [11:11] that people tell me you have to move [11:12] away from Visual Studio Code is [11:14] basically a fork of Visual Studio Code [11:16] with all the security settings turned [11:18] off because we want to go to the [11:20] terminal and and create random files [11:22] without you knowing what that we do. And [11:25] most of them actually are like wind surf [11:27] whatever AI features. 50 free tools cost [11:30] you tokens every single time you do [11:32] something. And when you do the math, on [11:34] average, this means 100k for a [11:36] development team a year, a small [11:38] development team. For an agency, that's [11:41] a lot of money to come in just to get [11:42] your developers to code. And I'm like, I [11:45] didn't want to buy at Dreamweaver. I [11:47] didn't want to buy any IDE. I always [11:49] used open source idees because I want to [11:51] be a free developer and I don't want to [11:53] pay to code. And this is a hidden way to [11:55] actually pay to code. So it is a [11:57] subscription in disguise to a degree and [12:00] the whole story that AI is making [12:02] engineers 10 times more productive is [12:04] not true. It's true sometimes for like [12:06] things people that have to build MVPs [12:08] for a living but the uh basically this [12:11] is a lot of links that I that came from [12:13] the newsletter and the developer survey [12:15] from Stack Overflow shows for example [12:16] the trust in AI coding tools is falling [12:18] and the usage rises. So, [12:21] a lot of developers that I talk to get [12:23] forced to use AI in their job now [12:25] because the company thinks they're [12:27] falling behind if they're not using AI, [12:29] but it's really not helping them because [12:30] the code hasn't changed in 10 years and [12:32] there's not much cool stuff to build. [12:34] It's just things to maintain and LLMs [12:36] are bad at maintaining things because [12:37] they just keep rewriting them and [12:39] starting them a new. [12:42] And there were interesting insights in [12:43] that data in that in that survey and [12:46] that's basically that uh 33% of [12:48] developers trust AI accuracy in 2025 43% [12:52] did that in 2024 and 42% in 2023. So it [12:56] goes down and I feel the same. The [12:58] accuracy of of code completion sometimes [13:00] is really odd especially when you do [13:03] languages that you don't normally use or [13:05] you just basically sometimes I use Pearl [13:07] and people like what's wrong with you? [13:08] And these kind of things AI favorability [13:12] dropped from 77% to 72 and 60% now. So [13:17] people are just like that [13:18] disillusionment starts to get in. People [13:20] are starting like it's not really as [13:22] useful as it is. And 66% say AI [13:25] solutions that are almost right but not [13:27] quite is the top frustration. And that's [13:30] the same when you had a code review with [13:31] your colleagues, right? You're like, [13:33] "Okay, I forgot about that one thing." [13:34] And you're like, and the LLM is all [13:36] like, "Oh, apologetic. I start from [13:37] scratch again." You're like, "No, no, [13:38] no. Just fix that one thing." Like, "No, [13:40] don't don't start from scratch again." [13:42] But yeah, I'm sorry. Please use more [13:44] tokens. Uh, I mean, it's great to be a [13:46] VI coder because by by lunchtime, you [13:47] have used up the tokens for a day and [13:49] you can go home. [13:51] 45% say debugging AI generated code [13:53] takes more time than expected and AI [13:55] tools promise productivity but actually [13:57] create new categories of technical debt. [14:00] And writing or generating code should [14:02] always aim to solve an issue. I've never [14:04] been excited about writing code. I've [14:06] been excited about deleting code that's [14:08] not necessary anymore because standards [14:10] became better. And I've been excited [14:12] about solving problems with code. [14:13] Sometimes horrible code that nobody can [14:15] read but me. But that's my problem. But [14:17] it's not it's all it's always about the [14:19] issue not like how much code we write. I [14:22] mean when certain person bought a [14:24] certain chat system and then basically [14:26] said like I measure my engineers by how [14:28] many code how much code they write a [14:29] month. I could have written them so much [14:31] crap like you know just like wait for [14:34] loop that's could be 50 if statements [14:36] that's 50 more lines done like so VIP [14:40] coding has always been to me like this [14:42] uh uh put the thing in and make it [14:43] happen. The thing that irked me the most [14:45] about the original paper of VIP coding [14:47] was not that he that he said like don't [14:50] forget forget about the code just think [14:52] about talk to your machine and the [14:53] machine will do it for you. I actually [14:54] like that idea. What I hated is that he [14:56] wrote a game that he can put ads in to [14:59] make money with the pro the aim was a [15:02] product to make money not write a code [15:04] or solve a code problem. Not like I want [15:06] to learn how to write a game so the AI [15:08] can help me do that game. No, I want to [15:10] show I want to show ads to people. And [15:12] I'm like, this has been Yahoo in 2001. [15:15] We've done that. That's nothing new. [15:16] There's nothing clever about this. I [15:19] like that article that was talking about [15:20] VIP code as legacy code. And he [15:23] explained that like code that nobody [15:25] understands is tech. It takes a lot of [15:27] time to understand unfamiliar code [15:29] enough to debug it, let alone introduce [15:31] new features without also introducing [15:32] bugs. So the whole idea of like don't [15:34] think of the code that generates, think [15:36] about the next feature. Think about that [15:38] cool thing. uh when you vip code, you're [15:40] incurring tech debt as far fast as the [15:42] LLM can spit it out. This is why VIP [15:44] coding is perfect for prototypes or [15:45] throwaway things. That's why I love it [15:47] as well. But legacy code as soon as you [15:50] put it out there and maintain it. And [15:52] nobody likes writing MVPs or or like [15:55] like little prototypes or like slideware [15:57] or things like that, that's do that with [15:59] LLMs. Fine. Like throw the thing away. [16:02] But then again, do we really need to? We [16:03] can do it with a paper prototype as [16:05] well. like it's not really that [16:07] important to build an app for people to [16:09] click on. I have a new feature coming in [16:12] my podcast soon about um [laughter] [16:15] about confessions, tech confessions. And [16:17] my best story that I have about this was [16:19] in 200 no 1997 in England uh 1999 in [16:24] England. Uh I had an energy provider [16:26] that we built the website for them [16:28] £20,000 or something like that that not [16:30] not that high but and he said yeah well [16:33] we can't we can't pay you or put it live [16:36] because it doesn't look right like good [16:38] thanks for this detailed feedback that [16:40] really helps me a lot. So uh I'm coming [16:42] to the office and let's talk about it. [16:43] So I came to his office and he had [16:45] Internet Explorer 5 O2 because I asked [16:48] him like why didn't you upgrade to [16:50] Internet Explorer 6? It's about half a [16:51] year old. Oh, Internet Explorer 6 has [16:53] all these viruses so I can't trust it. [16:56] So I knew I can't talk to that person. [16:58] But that was before Wi-Fi and it was a a [17:02] wire connection. So I checked his IP [17:05] number and we showed him a screenshot of [17:07] the website rather than the website for [17:09] half a year and he always signed off [17:11] every month and paid us for it. He never [17:14] clicked on the thing. He just wanted to [17:15] see the first render look right. So if [17:17] you can get away with that, get away [17:19] with that. But like yeah, it's on. [17:22] And yes, CEOs can now do production [17:24] code. That's the great thing about VIP [17:26] coding. I love the the first outage from [17:28] LLM written code by sketch.dev def and [17:31] it was like the trigger for the CPU [17:34] spikes that crashed the server both [17:36] times was our CEO logging in. We [17:38] redeployed to get a clean start, [17:40] permanently banned him from the service [17:41] and moved on. [17:44] I think this is a good idea generally [17:46] like you know that's not a bad plan but [17:48] the my favorite was what the LLM did. Uh [17:51] this was the code that actually was a [17:53] for loop and it basically had an if [17:56] error is not nil. I don't even know what [17:57] that language is. Uh and then break of [18:00] course if error is not nil break. What [18:02] did the LLM do? Because it wants to be [18:04] nice and happy and good. It put changed [18:07] to break to a continue. [18:09] So the endless loop kept piling up and [18:11] that crashed the whole thing. [18:14] Now how do I use AI right now? And [18:18] that's something that I found very [18:19] interesting. Uh we have a ton of [18:22] content. I got this weekly newsletter. [18:24] I've got a weekly live show. I got a [18:25] magazine articles podcast. 10 years of [18:28] we developer world congress that means [18:29] like 2,500 to 3,000 hours of talk and [18:32] workshop footage times 10 and last year [18:35] it was 12 terabyte of data of videos [18:38] that I had to postproduce [18:40] did a 50 line PHP script and ffmpeact to [18:43] do the thing but like it's it basically [18:44] was a lot of data to be shifted around [18:47] and I was wondering like what can we do [18:48] about making it easier to search things [18:50] and make it more interesting for you to [18:53] use that data and reuse that data later [18:55] on as well. One tool I use is video tap [18:58] uh which actually allows you to upload a [19:00] video and then it generates like shorts [19:01] automatically by finding interesting [19:03] things in the in the in the video uh [19:06] creating like tweets creating podcast [19:08] things and everything for it and that we [19:11] then thought okay we can automate that [19:13] on the server side as well in our own [19:15] thing. So in our website right now [19:17] there's three things that we started to [19:19] build and that is the video button, a [19:21] moments button and the playlist button. [19:23] The video button is just as you would [19:25] expect in any other platform. You just [19:27] search for a topic and you get videos to [19:29] actually about that topic that has it in [19:32] there. We create the transcript. We scan [19:34] through the transcript and then [19:35] basically tell you which videos is [19:37] rather than tagging it by hand. [19:40] The moments was more interesting because [19:42] we then give you all the videos that [19:44] talk about a certain subject matter for [19:46] example and give you the times in those [19:48] videos where that person or that talk [19:51] covered that topic. Again going through [19:53] the uh going through the transcript and [19:55] also going through the audio file and [19:57] finding it like that. And then we [19:59] thought of course the playlist would be [20:00] then we give you not the full videos but [20:02] only those sections of the videos and [20:05] you can have that running in the [20:06] background and you have like now the [20:08] playlist for security best practices as [20:11] little audio files and as an in-built [20:13] thing in the browser running in the [20:15] background. Next step now will be to [20:17] generate those videos for you that you [20:19] can download them if you wanted to as [20:20] well. But I found it very fascinating [20:22] how easy it is by tra by analyzing the [20:24] transcript and then uh getting into the [20:27] stream of the video. The only thing we [20:28] found is latency. YouTube was absolutely [20:31] atrocious in latency. Vimeo as well. So [20:33] we're now going for Cloudflare for the [20:35] hosting of that one. So it might be [20:37] might be easier to do that. But I love [20:39] that we can now give you like videos [20:40] from four years ago that are still [20:42] relevant, but you would have never known [20:43] them because nobody watches a video [20:45] that's longer than 10 minutes. And this [20:47] is helps you with that stuff. [20:49] Now let's talk about our jobs right now [20:51] and let's talk about how it actually [20:53] affects uh uh the the the careers that [20:56] we have in that AI world where people [20:58] see code as a throwaway thing and the [21:00] app to be the thing that you should be [21:02] taking for. Now you probably all seen [21:05] that like we used to do waterfall things [21:07] where we basically like okay the [21:09] software just would not work until we [21:11] actually go through the last stage. So [21:13] you start with the wheel and then you [21:14] put the the chassis on it and then you [21:16] get the car. And then of course we told [21:18] people agile is a better idea. Just [21:19] build things that keep working and make [21:21] them faster and release a new working [21:23] product every day. Now with the with the [21:26] code being out of the picture and the [21:28] LM's doing it for you in the AI world, [21:31] we basically have this instead [21:34] cuz you know instead of just building [21:36] something that works, we build something [21:37] that kind of works every single time in [21:39] different different ways. And to me that [21:42] goes back to the the biggest step [21:44] backwards in software distribution. And [21:46] I gave a TED talk about this TEDex not [21:48] TED um uh which was apps. I hated when [21:53] app stores came out because it took away [21:56] the ownership of software from the user [21:59] to the provider. So back in the days you [22:02] had like floppy discs, CDs, DVDs or to [22:05] get your software and basically uh you [22:07] when it broke you're done. You have to [22:09] buy the thing again. Then the internet [22:11] came and I'm like, "Hey, new version [22:12] upload. I can do I can change the [22:14] thing." Wonderful. Great. Then the app [22:16] store came and people say like, "Well, [22:18] that software cannot be released in that [22:20] country anymore. So, it's not going to [22:21] be available in the in the app store. [22:24] You cannot delete stuff from the [22:25] internet, but you can delete things from [22:27] app stores." And I worked on the app [22:30] store in Microsoft. I worked on the app [22:31] store in Mozilla. I worked in the app [22:32] store in Yahoo. None of them bloody [22:34] scale. None of them work. And none of [22:36] them give the more give us more security [22:38] as the people say as well. like the the [22:40] amount of malicious VPNs on the iOS [22:43] store and the Android store out there. [22:45] When Sora just came out, there's there's [22:47] like 20 no 72 cloned Sora apps that are [22:51] not the Sora thing, but actually full of [22:53] malware and that was in the Android [22:54] store within 3 minutes. It's the same [22:56] thing, but it's just more control of the [22:59] publisher rather than the the you user. [23:02] The other thing is like I learned a lot [23:04] from making mistakes and we do that. We [23:07] learn by imitating and by making [23:09] mistakes. And how does that actually [23:12] translate for junior developers now? [23:14] Like how can we say that junior [23:15] developers learn if they're encouraged [23:18] to let magical AI do everything for [23:20] them? Like they never get a chance to [23:22] build the thing. It always cracks me up [23:24] when everybody talks about developer [23:26] efficiency and like how it makes us more [23:29] more more faster and we have more time [23:31] to do what? I good give your engineers [23:34] the things to automatically do but then [23:36] give them time to learn to improve to [23:39] share with each other to build something [23:41] internally that wasn't done before [23:43] internal hack days and these kind of [23:44] things but the time we save by using [23:46] LLMs we never actually push anywhere [23:49] else except for using more LLM things [23:51] it's it's an addiction machine it's the [23:53] same as like going Tik Tok or like [23:55] Facebook and these kind of things I [23:57] talked about that in the keynote that [23:58] basically the Dunning Kruber effect the [24:00] cognitive bias that people think They're [24:02] engineers, they're designers, they're [24:04] they're musicians, they're creative [24:06] people just by using AI [24:08] is dangerous to me because let's all [24:11] let's all not forget it's uh it's just [24:14] bad for us if we if we think we're much [24:16] better than we actually are. But it's [24:18] also all costly. [24:20] One image generation is the same amount [24:23] of energy than charging your mobile [24:24] phone. Was it really worthwhile having [24:27] that turtle on a skateboard that you [24:28] send your friends for one smile and one [24:30] like? Do we actually need this? Do we [24:33] want to use technology that that needs [24:35] more energy? I I would have loved if by [24:38] now we would have solved like global [24:39] warming or something like that. One [24:41] thing that that the co bit taught me is [24:43] that we're all bullshitting. [24:46] Google IO other keynotes always years [24:48] ago always had like, oh, we're going to [24:50] have contact lenses that can that can [24:52] save save you from cancer. We're going [24:54] to have self-driving cars that clean [24:56] your garden or whatever. You know, [24:58] everything was about like we solve that [25:00] problem. We solve that problem. And then [25:01] COVID came, everybody was shocked like, [25:03] "Oh, we don't know about that disease." [25:04] And then nobody managed to build an app [25:07] that tracks who's been vaccinated or [25:09] not. That's a bloody spreadsheet. That's [25:11] not that hard to do. But the German [25:14] government spent 10 million euros on an [25:15] app that everybody had on their phone [25:17] that said like, "Here's your vaccination [25:18] status and here's how far away you from [25:20] somebody who who just had an outbreak of [25:22] of COVID." [25:25] Two years later, when it was over, it's [25:26] not over, but when it was over, they [25:28] just let the thing delete. it just [25:30] deleted anymore. You built an [25:31] infrastructure. Everybody had an app on [25:33] it. I should do my taxes on that thing. [25:35] I should do my doctor appointments on [25:36] that thing. But like, no, no, co is [25:38] done. Did we do something new now? And [25:40] it was just fascinating to me that like [25:42] every technology company failed to solve [25:44] that simple problem. And at the same [25:46] time, we're now back at like, yeah, [25:48] there's going to be a robot that does [25:49] your yesterday there's going to be a [25:51] robot that does your dishes and folds [25:52] your your shirt your shirts and stuff. I [25:54] don't fold shirts. I don't need a robot [25:56] for that. [25:59] Talking of horrible shirts, I think he [26:00] lost uh um he lost a bet back then and [26:04] that was my favorite moment in the whole [26:05] thing. Like the entrepreneur of the year [26:07] uh uh basically builder.ai turned out to [26:11] be like 700 developers in India. It was [26:13] not an AI at all and they did a lot of [26:15] moneyaundering and all kind of things. [26:17] It was one one of our and it was one of [26:19] those unicorns like 1 billion funding [26:21] and it was complete horshit from [26:22] beginning to end. And this is yeah what [26:25] happens. This is this is still possible, [26:27] but I hope that we're getting past that [26:30] and think like how can you use AI in a [26:32] sensible fashion that makes more sense [26:34] to people. We protected the web from [26:36] corporate takeover before and we need to [26:38] do this again. I need you from that. I [26:40] need everybody else from that. I need [26:41] new developers coming into the market [26:43] and demanding better and demanding not [26:45] to have to pay to play and be in it. But [26:48] that your skills, your ideas should be [26:50] the currency to make you a developer in [26:53] certain companies. [26:54] And also um [26:58] I think it was Windsurf when they got [26:59] acquired and then the the the new CEO of [27:02] the other company that bought them was [27:03] like well now we want you to be in the [27:05] office 70 hours a week because we love [27:07] software. No, you hate people. [27:11] There's a difference between the two. [27:12] Like after 50 hours we're done. We're [27:15] cooked. We only make make nonsensical [27:17] things or we make things that think [27:19] we're good. Like whenever you go on a [27:21] plane and you watch a movie and you're [27:22] like that was the best movie ever and [27:24] then you go back home and you watch it [27:25] again you're like what the hell was I [27:26] thinking? But that's the same when [27:28] you're basically tired when you when you [27:30] have worked too long. [27:32] We should be in a place where I can work [27:35] 30 hours well and then go home rather [27:38] than like oh it's so cool to to do the [27:40] side hustle and have like five more [27:41] startups and these kind of things. [27:43] That's like investment bankers in the [27:44] 80s. They hate I hated it then. Don't do [27:46] that. Like it makes me cringe to [27:48] actually think that work uh uh workers [27:51] rights in our world is just basically [27:54] given away for like here's short lots of [27:56] money or uh or shares. My dad was a coal [27:59] miner and then a factory worker. So I [28:01] grew up with that whole environment. [28:02] People were on the streets for rights to [28:04] not work on a Sunday and basically yeah [28:07] nowadays yeah I work on a Sunday because [28:08] I have a side hustle. No, don't do that. [28:12] Not by demonizing AI and companies that [28:14] succeed in it like just I did. Okay. But [28:16] by taking part in offering an [28:17] alternatives and AI can make us more [28:19] efficient. It can't make us more [28:21] creative or magically excellent at [28:22] anything. The whole idea of like anybody [28:25] can be creative. And I mean our mobile [28:27] phones now like these lenses don't make [28:29] a difference anymore. Like the picture [28:31] as soon as you take it there's software [28:33] running over it and it makes the picture [28:34] better than it was. And it feels like [28:37] you're a great photographer but actually [28:38] the machine does most for you. And I [28:40] kind of miss that errors. I miss these [28:43] things where it's where where things are [28:44] wrong. Like all the pictures in the 80s [28:46] where people have these red eyes. If if [28:48] any kid asks you what that is, says like [28:50] it was a demon possession for three [28:51] years, but we don't talk about it [28:52] anymore. [28:56] So let's show the world how the power, [28:58] love, and dedication of open can move [29:00] this from a marketing hype to the next [29:02] level of human computer interaction. I [29:04] want computing to be ubiquitous. I want [29:06] computing to be possible for me [29:08] everywhere. But I don't want anything [29:11] that gives me one answer and I have no [29:12] idea where it came from. That's the [29:14] thing that I find weird about it. So if [29:17] you want to help me, the the World [29:20] Congress 2026 is actually happening on 8 [29:23] to 10th of July. It's free for you if [29:27] you get picked as a speaker. [29:30] [laughter] [29:32] Uh we also need volunteers, but I I [29:34] guess that's kind of like pointless for [29:35] here because you're not you're not here. [29:37] But we have the one in San Jose that [29:39] might be easier for you. You only have [29:40] to fly to the other side. Uh 10% cheaper [29:43] for all things open people and free if [29:46] you get speak picked as a speaker here. [29:49] And I'm far too fast, but I I think that [29:52] was kind of it. So, thanks very much.