How to read this
📰 Quick hits
The headlines worth knowing even if you read nothing else this week.
via Claude Team / Hacker Newsletter
via Claude Team
OpenAI acquires Ona (formerly Gitpod)
via The Pragmatic Engineer
Open models now sufficient for ~60% of coding work
via The Pragmatic Engineer
Ford rehires "gray beard" quality inspectors after AI fell short
via Hacker Newsletter
via Words That Matter
🧠 How tools reshape cognition
The questionWhat does using AI actually do to how we think, learn, and pay attention?
There Are Three Types of AI Users
theatlantic.com (David Brooks)
David Brooks argues the differentiator in the AI age isn't intelligence but your relationship to mental effort: the paradox of tools that promise to free thinking while eroding the capacity to think. (Atlantic headline: "The People Who Will Thrive in the AI Age".)
"What will differentiate people is not how smart they are but their relationship to mental effort."
Expertise in the Age of AI
moderndescartes.com
Uses the vanished human "calculator" job as an analogy for coding agents: the tool removes mechanical work but makes the underlying intuition more valuable. Skipping the struggle leaves you unable to judge the output.
"Doing the work is the best way to build mastery... Don't use AI until you've done it by hand at least once."
The right kind of AI sceptic
theengineeringmanager.com
Distinguishes scepticism-as-conclusion from scepticism-as-identity; the real test is whether you can say what would change your mind. The author turns the knife on his own enthusiast position.
"The enthusiast who dismisses every failure as 'early days' is doing the same thing as the sceptic who dismisses every success as 'cherry-picked.' Neither is thinking clearly, and both are optimising for being right over being accurate."
Air Traffic Control
newsletter.kentbeck.com
Kent Beck and systems veteran Keith Adams on what happens when the twenty-year software playbook goes blank: Jevons paradox, compute-as-moat, and the flow state of programming traded for something closer to air traffic control.
"the flow state that drew us to programming, now traded for something closer to air traffic control."
🔍 Translation vs. understanding
The questionIs AI genuinely understanding, or just translating context into plausible output, and where does real human comprehension still earn its keep?
The twilight of the chatbots
Ethan Mollick · One Useful Thing
Mollick's structural read on the shift from co-intelligence chatbots to assigning work to agents: the operator's job is now management, and domain expertise (not profession) predicts how much useful output you extract.
"What actually mattered was not the profession of the user, but their expertise... We are moving from a world where non-experts use chatbots to fill in gaps to one in which experts use agents to get work done."
How Kent Beck shapes the software engineering industry
newsletter.pragmaticengineer.com
A career-spanning retrospective (TDD, XP, Agile) re-examined for the AI era, with the structural claim that we're failing to accumulate trust at the rate we accumulate code.
"your ability to affect change in the world is gated by your ability to communicate with, to soothe, to understand other human beings. And those are exactly the skills that I thought I didn't need to learn!"
Please stop the AI Confidence Theater
Elena Verna · Lenny's Newsletter
Elena Verna complicates the AI-hype consensus with a "Show me" heuristic and a sharp hiring point: AI made average intelligence cheap and handed everyone the vocabulary of expertise, so sounding competent and being competent now diverge.
"AI has made average intelligence incredibly cheap and it has given everyone the vocabulary of expertise... But sounding competent and being competent are completely different things."
A coding agent is six functions in a trenchcoat
tidydesign.substack.com
Hadley Wickham demystifies coding agents by building a minimal one from three tools, then layering in safety and edit tools. A build-it-to-understand-it deep dive.
"The shell tool is our 'get out of jail free' card because if you can run a shell command you can do anything... The shell tool is also rather dangerous; we'll come back to that later."
Working With AI: A Concrete Example
htmx.org
Carson Gross walks through a concrete hyperscript parser bug fix to show AI's strengths (investigation, tests) and its weakness (clean architecture), narrowly avoiding the "sorcerer's apprentice" trap because he understood the codebase.
"rather than being a sorcerer's apprentice and blindly accepting the solutions AI proposed, I was acting as a sorcerer... demanding a correct solution that better fit the existing codebase's architecture."
💰 Value concentration when creation costs collapse
The questionWhen building something gets cheap, where does the value (and the money) actually pool up?
Aesthetic Warfare (and the Power of Aesthetic Authorship)
Anu Atluru · Words That Matter
When AI can execute any technique flawlessly, value concentrates in aesthetic authorship: the who and why behind a coherent body of work. Cross-domain analysis across art, fashion, film, politics, and business. (Listed in the email as "The Aesthetic Is The Art Now".)
"AI is great at writing, and even good at the entire pyramid of execution agency — except for the very tip of agency which is actually making creative decisions about what's worth writing at all and why... This is almost the essence of writing."
New Media Is Insider Media
Anu Atluru · Words That Matter
A structural theory that a small audience of the right people beats a mass audience, because prestige converts judgment into influence and judgment is independent of scale.
"Insider Media helps institutions that are smaller than their rivals become culturally larger than them... it converts judgment into influence—and judgment is independent of scale."
Impressions from visiting OpenAI, Anthropic, & Cursor
newsletter.pragmaticengineer.com
Firsthand structural analysis of cloud agents going mainstream and engineering work shifting toward "building environments for agents to execute efficiently." Where does engineering value migrate when execution is automated?
"Agents in the cloud don't have a way to 'complain.'... Cursor came up with the idea for the model to 'confess' in regular interviews."
Local Reasoning for Global Properties
tratt.net
Laurence Tratt visibly changes his mind on whether AI needs new languages: AI is strong locally but weak globally, so languages that enforce global properties via local reasoning (Rust's data-race freedom) regain value.
"If you think a program can be in states A, B, and C... but two of those states cannot happen, you've not just wasted effort, but created a potentially exponential explosion of states for subsequent edits to consider."
You can't unit test for taste
karltryggvason.com
A candid retrospective building a real points-of-interest pipeline where AI got demoted from writing (it hallucinated) but promoted for the subjective taste latent in its weights, and where verification had no ground truth.
"Verification becomes hard to reason about because there is no ground truth for points of interest, there are no red/green unit tests for taste."
🪵 Thick engagement vs. thin optimization
The questionWhen is the slow, effortful, deep version of the work worth it, versus the fast and frictionless one?
Make Something Heavy
Anu Atluru · Words That Matter
Argues creation is a process of becoming where the work transforms the maker, treating frictionless AI output as a structural problem rather than an "AI is the future" opinion piece. Squarely on the question of what to hold onto as execution cost approaches zero.
"It was all motion, no mass—momentum without weight. 99% dopamine, near-zero serotonin, and no trace of oxytocin. This is the contemporary creator's dilemma—the contemporary generation's dilemma."
"My uncle used Claude to write my Nana's obituary"
Substack Reads (post.substack.com)
A first-person reckoning with what should be done by hand versus offloaded to AI when producing words costs nothing. Lands on the core question of what humans hold onto as execution becomes free.
"But the thing about words after somebody dies is that they are not for the person who is dead. They're for us, the living... knowing damn well that it is for me, knowing damn well that I am alive."
RDEL #150: How is AI adoption affecting team motivation?
rdel.substack.com
Research-driven analysis (LeadDev 2026 survey) of why engineer motivation erodes as the job shifts from authoring to reviewing, from creating to verifying. Complicates the productivity-only narrative.
"One response characterized a 'TikTok-ification' of problem solving that turned engineering into 'a babysitting chore, rather than doing interesting work.'"
🚀 Small teams, disproportionate output
The questionHow do tiny teams punch so far above their weight?
How Gusto Built a New Product Line in 10 Weeks with Claude Code, No Jira, and No Docs
How I AI · Lenny's Newsletter
A five-person team shipped a tier-one product in 10 weeks by stripping process to a perma-Zoom room and Claude Code, with the PR replacing the PRD. Coordination overhead becomes optional when the model carries state. (Also hits Planning Artifacts.)
"We had no meetings. We had no text specs. We had no Figma. We had no Jira board... We had no standups, no retros. We had nothing."
Software Factory for Agent Tools (ai that works #63)
Boundary · ai that works
A working "software factory" where agents write BAML around the clock, capture failures as "trophies," file tickets, open PRs, and answer CI; humans only decide what counts. A warts-and-all practitioner build.
"The humans hit merge. That's the whole thing... You do not have to build the whole factory. The redraft loop is just: agent drafts an issue, human adds a comment, agent redrafts."
A return to two-pizza culture
allthingsdistributed.com
Werner Vogels on how coding agents invert Amazon's write-first "Working Backwards" process into prototype-first, and reaffirm small autonomous teams. (Also hits Planning Artifacts.)
"You will learn more in one evening of building than in two weeks of writing about what you think will happen... The document you produce after building is fundamentally better... because it's no longer grounded in your assumptions."
🗺️ Planning artifacts shape the work
The questionHow do the documents you write (specs, decision records, the agent's workspace) steer what actually gets built?
How to Write an Effective Software Design Document
refactoringenglish.com
A practitioner-grounded treatment of design docs as the artifact that articulates hard problems and enables feedback before code is written, with a sharp heuristic for deciding what belongs in the plan.
"If you specify every possible detail in a design doc, you've essentially written the implementation during the design phase... ask a simple question: what's the penalty for being wrong?"
👀 On the radar
Lower-confidence picks worth a skim if the topic grabs you.
- GLM 5.2: A Live Review of an Opus-Level Open-Weights Model An unscripted practitioner test of an open-weight model inside a real codebase, framed as cost/control/vendor-lock-in rather than capability ceiling (a full 45-min agentic session cost $3.36).
- How top PMs increase their leverage with AI (the "ladders of leverage") Colin Matthews' three-ladders framework (personal / product / systems leverage) with concrete MCP + Claude Code prototyping patterns for non-engineers. PM-specific and marketing-heavy, but the judgment point on when to write a doc vs. ship a prototype is useful.
- Introducing the Tolaria Alliance (solo-dev agentic workflow: Guides, Gates, Guards) Luca Rossi's solo open-source product run largely by AI agents, steered via Guides (AGENTS/skills), deterministic Gates (commit-quality gates), and Guards. Substance is lighter and sponsorship-wrapped.
- Attention is all we have David Bessis frames attention and the focus of curiosity as the one thing that is truly ours; a conjectural theory of cognitive inequality drawing on how mathematicians develop through attention control. Adjacent to how systems shape attention.
- What Is Your Job Now? (Farhan Thawar, Compile 26) Shopify's head of engineering argues the coding bottleneck, once AI solves it, just relocates to planning, validation, and review (Goldratt's The Goal). A clean outside-in confirmation that the constraint moved, even if it stops short of rethinking the unit of work itself.