AI Is Devaluing Code; an Engineer's Moat Is Taste and Accountability
Addy Osmani, who leads Chrome DevTools, argues that AI has devalued writing code and that an engineer's real edge is taste and accountability — and warns against cognitive surrender, treating the AI's answer as your own.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
The hard part of a closing keynote is not what to say but whom not to repeat
Addy ran multiple agents to gather every talk, summary and social post he could find from the preceding days of the AI World Fair, and used that to sculpt his own closing keynote so it would not repeat points other speakers had already covered in depth. He says this kind of thing used to be nearly impossible, or would have taken an enormous amount of time; now it is entirely within reach. He describes running large parts of his daily life through agents as a coexistence of «fun and chaos».
— AddyEvery framework claims to be simpler; only the same app can settle it
In the era when Angular, Backbone and YUI were all emerging at once, Addy noticed that every framework claimed to make development simpler, yet there was no way to compare them directly. So he built the same to-do application, implemented it in each framework with a standardized feature set — that is ToDo MVC. The project quickly picked up thousands of stars and became the default example in many framework tutorials. Later, Apple's Safari/WebKit team reached out and they jointly developed Speedometer, which today is the responsiveness and web-application benchmark maintained collectively by all the browser vendors.
— AddyPage load is not one metric but three moments for the user
The Chrome team broke «page load» out of its vague form and into key moments in the user's journey: is it happening, is it useful, is it usable — mapping respectively to metrics like LCP, FID/INP and CLS. These metrics were not pulled out of thin air: they started from user research, then went through repeated validation with the standards community and with developers to confirm they really do correspond to how people judge a page's value.
— Addy OsmaniTreating the AI's answer as your own answer is cognitive surrender
Addy distinguishes cognitive debt from cognitive surrender: the former is the erosion of memory and comprehension that comes from long-term AI use; the latter is blindly accepting the AI's answer, treating its answer as your own, and gradually giving up critical thinking. He argues for avoiding that state — otherwise, when the agent gets something wrong, the engineer can only pray. To keep his own understanding intact, he insists on reading the final decision summary, and if there isn't one he asks the model to produce its decision process, while staying alert to the possibility that a model fabricates content when its context window is nearly exhausted.
— Addy OsmaniAn engineer's alpha is taste, and every model release shrinks it
Addy defines alpha as «edge», and says it decays with every model release. The software engineer's current alpha is taste: judging whether you are building the right thing, where to put your energy, and whether what you built is actually «good». An agent can judge whether something matches a spec, but has a hard time judging «good» — including whether the experience is delightful and whether users want to come back. Even if future models close that gap, engineers will still be accountable, the way Chromium's OWNERS files make them accountable, for understanding the system and for gating and shipping decisions.
— Addy OsmaniNeatly structured writing may lose to rough, real writing
Video creator Navi (Neat Code) publicly criticized Addy's long piece on loop engineering as «abstract, like it was written by AI». Addy acknowledges the piece was a first draft he wrote by hand, edited himself over several rounds, and finally handed to a model for a readability pass. He has always favored structured writing, but he is starting to suspect that what readers actually want may be rawer, imperfect writing that feels real. He also mentions Michael Novati: the piece that got criticized as «AI-like» was in fact one where he had used AI heavily; when he switched to multiple rounds of manual tweaking of the output, the result read noticeably better.
— Addy OsmaniDon't just be an engineer: the job boundaries are being unbundled
Addy expects an «unbundling» of engineering, product and UX roles: engineers will need product sense, product people will need engineering or UX sense, and the boundaries between roles will blur. His advice to anyone preparing to make a move is not to stare only through the narrow lens of engineering, but to understand product, developer advocacy, marketing and the other pieces of commercial success. Show employers you are not just a builder but someone who can adapt to blurry boundaries. The core of it is one line: Don't just be an engineer.
— Addy OsmaniIn their own words · checked verbatim
So agents have allowed me to take the improbable and turn it into the possible in ways that are kind of weird and wonderful every day.
Addy2:05
And the natural follow up to that is cognitive surrender, which is where you, you know, you blindly give in to whatever the AI says as your answer. Its answer becomes your answer. And so you start to really let go of critical thinking and your ability to solve problems just goes to the wayside.
Addy Osmani1:00:48
So for software engineers, we very often say that your alpha is in taste in terms of, are we building the right thing? Where are we putting our energy? Is the thing that we are building actually good?
Addy Osmani1:12:16
It's about forming a thesis about what people are struggling with and what educational content could be useful for them around that.
Addy Osmani1:19:28
And I feel like I easily wasted three hours of time because no matter what I did, one of the models I was using continued to generate text that looked like it had triads and these patterns, you know, patterns of AI written text.
Addy Osmani1:20:33
The more capable AI agents become, the easier it is to let them do what they do.
Host1:29:53
Figures
| Years at Google | 14 years | 44:30 |
| Level when he joined Google | L4 | 51:01 |
| Average size of the team he manages | 45-50 people | 52:08 |
| Books published | 18 | 1:18:26 |
| Time invested in writing one article | at least 3-7 days | 1:24:44 |
| Time wasted on readability tweaking | 3 hours | 1:20:33 |
Glossary
- cognitive surrender
- The state of blindly accepting the AI's answers and giving up critical thinking.
- alpha
- An investing term for the ability to beat a benchmark; Addy borrows it for what an engineer uniquely offers relative to AI.
- unbundling
- Skills once packaged into a single role get pulled apart, blurring the boundaries between jobs.
How to listen
Frontend engineers, engineering leads, AI product managers, and any developer worried about being replaced by AI.
You can fast-forward the first 14 minutes of childhood backstory; the methodology section after minute 46 is the part to listen to.