--- title: Let the model cook, but own what ships | daily.dev description: Addy Osmani's AI Engineer keynote argues that as developers delegate more to AI agents, their accountability increases rather than decreases. He introduces... canonical: https://daily.dev/posts/let-the-model-cook-but-own-what-ships-ycssydoyo twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Let the model cook, but own what ships | daily.dev og:description: Addy Osmani's AI Engineer keynote argues that as developers delegate more to AI agents, their accountability increases rather than decreases. He introduces... og:url: https://daily.dev/posts/let-the-model-cook-but-own-what-ships-ycssydoyo og:image: https://api.daily.dev/og/posts/yCSsYdOyO.png og:image:alt: Let the model cook, but own what ships og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Let the model cook, but own what ships **[Trends](https://daily.dev/sources/trends)** · 2 min read · 3 upvotes · 1 comments ## Summary Addy Osmani's AI Engineer keynote argues that as developers delegate more to AI agents, their accountability increases rather than decreases. He introduces three concepts: cognitive debt (losing understanding of agent-written code), cognitive surrender (accepting AI output without verification), and orchestration tax (the gap between parallelizing agents and parallelizing human judgment). His core rule is 'explain it or don't ship it' — not a productivity argument but an accountability one. He also frames this as a career question, distinguishing 'alpha' skills (edge now) from 'decay' skills (absorbed by models), arguing that accountability and credibility outlast speed and recall. The takeaway: let agents do the work, but the outer loop of deciding, verifying, and owning outcomes remains the developer's responsibility. ## Content There's a study making the rounds that should make every developer uncomfortable. French and Italian researchers ran an experiment where AI access made people **three times less accurate** and **twice as confident**. Willingness to say "I don't know" collapsed from 44% to 3%. The kicker: the researchers deliberately used questions where the AI typically fails, so this wasn't rational delegation. It was just... surrender. That word — cognitive surrender — is having a moment. Addy Osmani used it in his closing keynote at AI Engineer World's Fair, and it's landing because it names something developers are quietly noticing in themselves. Jerry Liu put it plainly: "I used to do this a lot. But then I felt like I was getting progressively dumber at writing." His fix is manual: type out your own understanding in bullet points after reading the LLM output. Force the brain to do something. Another developer noted the same creep in a different direction — "I'm starting to have trouble understanding and articulating what AI generates." Osmani's framing is the sharpest take in this whole conversation. His argument: agents should own the inner loop (investigate, implement, test, report), but engineers must own the outer loop — decide, verify, approve, own. The rule he lands on is blunt: **explain it or don't ship it.** He also introduces "cognitive debt" (you stop understanding your own codebase) and "orchestration tax" (parallelizing agents without parallelizing judgment). ThePrimeagen's read on all this is characteristically dark: "10x efficiency == 100x bad ideas." More capable harnesses just mean the dumb ideas ship faster. The counter-position, from Charity Majors and others, is that AI raises the bar rather than lowering it — you can ask dumb questions, prototype faster, learn new domains without embarrassment. "Why would you lower the bar for yourself by not bothering with any of this?" That's a fair point, but it assumes you're still engaging critically, which the research suggests a lot of people aren't. Kent C. Dodds put the stakes simply: if you're not asking questions while using AI agents, you're a button pusher, not an engineer. Curiosity is still the differentiator. The uncomfortable truth sitting under all of this: the skills with the longest half-life — accountability, judgment, the ability to explain a decision after the fact — are exactly the ones that atrophy fastest when you stop exercising them. ## Questions this post answers ### What did the study on AI and cognitive surrender find about accuracy and confidence? French and Italian researchers found that AI access made participants three times less accurate while making them twice as confident in their answers. Willingness to admit uncertainty by saying 'I don't know' collapsed from 44% to 3%. The questions were deliberately chosen from cases where AI typically fails, meaning the drop reflected uncritical reliance rather than reasonable delegation. _Engineers weighing how much to trust AI output can follow this kind of research on daily.dev._ ### What does Addy Osmani mean by engineers owning the outer loop with AI agents? Addy Osmani argues agents should handle the inner loop of investigating, implementing, testing, and reporting, while engineers must own the outer loop of deciding, verifying, approving, and taking ownership. His rule is blunt: explain it or don't ship it. He also coined 'cognitive debt' for losing understanding of your own codebase and 'orchestration tax' for parallelizing agents without parallelizing judgment. _Teams defining accountability rules for AI-assisted coding can track this kind of guidance on daily.dev._ ## Community discussion Top comments from developers on daily.dev. **@idoshamun** · 0 upvotes > The real taste comes when you ask questions and fine tune your agent. Not every task needs taste but when you do, you always need to steer the agent ## Similar posts on daily.dev - [Cognitive Surrender](https://daily.dev/posts/cognitive-surrender-ildbafhfw) · Addy Osmani · 171 upvotes · 12 comments - [Wharton researchers coined ‘cognitive surrender’ to describe what happens when people let AI think for them](https://daily.dev/posts/wharton-researchers-coined-cognitive-surrender-to-describe-what-happens-when-people-let-ai-think-f-zxqbhf5tb) · The Next Web · 1 upvotes · 0 comments - [The "I don't know, Claude wrote this" pandemic](https://daily.dev/posts/the-i-don-t-know-claude-wrote-this-pandemic-1gycwe8qz) · Manager.dev · 331 upvotes · 69 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#productivity](https://daily.dev/tags/productivity), [#vibe-coding](https://daily.dev/tags/vibe-coding) [View this post on daily.dev](https://daily.dev/posts/let-the-model-cook-but-own-what-ships-ycssydoyo) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Let the model cook, but own what ships","url":"https://daily.dev/posts/let-the-model-cook-but-own-what-ships-ycssydoyo","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/let-the-model-cook-but-own-what-ships-ycssydoyo"},"datePublished":"2026-07-15T16:50:32.487Z","dateModified":"2026-09-13T19:36:31.303Z","description":"Addy Osmani's AI Engineer keynote argues that as developers delegate more to AI agents, their accountability increases rather than decreases. He introduces...","isAccessibleForFree":true,"articleSection":"Trends","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Trends","logo":"https://media.daily.dev/image/upload/s--ZfSp3asX--/f_auto,q_auto/v1780996004/logos/trends?_a=BAMAMiWQ0","url":"https://daily.dev/sources/trends"},"commentCount":1,"discussionUrl":"https://daily.dev/posts/let-the-model-cook-but-own-what-ships-ycssydoyo","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":3},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":1}],"keywords":"ai-agents,productivity,vibe-coding","timeRequired":"PT2M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Trends","item":"https://daily.dev/sources/trends"},{"@type":"ListItem","position":3,"name":"Let the model cook, but own what ships"}]} {"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/let-the-model-cook-but-own-what-ships-ycssydoyo","comment":[{"@type":"Comment","text":"The real taste comes when you ask questions and fine tune your agent. Not every task needs taste but when you do, you always need to steer the agent","datePublished":"2026-07-16T18:12:28.584Z","url":"https://daily.dev/posts/yCSsYdOyO#c-2L1A50r5X","author":{"@type":"Person","name":"Ido Shamun","url":"https://daily.dev/idoshamun","image":"https://media.daily.dev/image/upload/s---xy_OAwk--/f_auto,q_auto/v1703781380/avatars/avatar_28849d86070e4c099c877ab6837c61f0"}}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/let-the-model-cook-but-own-what-ships-ycssydoyo#faq","mainEntity":[{"@type":"Question","name":"What did the study on AI and cognitive surrender find about accuracy and confidence?","acceptedAnswer":{"@type":"Answer","text":"French and Italian researchers found that AI access made participants three times less accurate while making them twice as confident in their answers. Willingness to admit uncertainty by saying 'I don't know' collapsed from 44% to 3%. The questions were deliberately chosen from cases where AI typically fails, meaning the drop reflected uncritical reliance rather than reasonable delegation. Engineers weighing how much to trust AI output can follow this kind of research on daily.dev."}},{"@type":"Question","name":"What does Addy Osmani mean by engineers owning the outer loop with AI agents?","acceptedAnswer":{"@type":"Answer","text":"Addy Osmani argues agents should handle the inner loop of investigating, implementing, testing, and reporting, while engineers must own the outer loop of deciding, verifying, approving, and taking ownership. His rule is blunt: explain it or don't ship it. He also coined 'cognitive debt' for losing understanding of your own codebase and 'orchestration tax' for parallelizing agents without parallelizing judgment. Teams defining accountability rules for AI-assisted coding can track this kind of guidance on daily.dev."}}]} ```