2026-04-03

this thread is what mass cope from legacy devs looks like.
i talked to @FastCompany about why @garrytan's "AI slop" is actually the future of software engineering.
the mass code review. the line-by-line gatekeeping. the "craftsmanship" that was really just slow iteration disguised as rigor - that era is over.
and the engineers who built their entire identity around it are panicking.
@gregorein brags about burning 3 billion tokens last year while dunking on garry for flexing lines of code. i've burned 6.6 billion in the past three months on codex alone. by his own logic, i'm 8x as credible.
see how silly that sounds?
yes, he found real issues. yes, they got fixed. that's exactly the point.
karpathy's autoresearch proved this already - AI agents can solve very complex problems just by operating inside feedback loops, iterating to optimize a loss function.
this is what software engineering is now - gradient descent.
ship, measure, self-correct, repeat.
all by the agent itself.
this is the new startup playbook.
your job isn't to review every line before deploy. your job is to build systems where agents observe outcomes - mrr, analytics, error rates, user behavior - and self-improve.
the engineer's role shifts from gatekeeper to building the machine that builds the machine.
you could run this level of audit (using AI) on any production site and find the same issues - most just don't have a billionaire CEO attached for virality.
mocking the people who adapted is easier than adapting.
but the craft is evolving whether you like it or not.
gregorein: so... I audited Garry's website after he bragged about 37K LOC/day and a 72-day shipping streak.
here's what 78,400 lines of AI slop code actually looks like in production.
a single homepage load of http://garryslist.org downloads 6.42 MB across 169 requests.
for a

We are truly just getting going on this one.
http://Every.to/plus-one
Ryan Boyle: I am so fucking excited to try Plus One...! Great onboarding @every

Charter reform his elected position so the mayor can appoint a competent Public Defender; which all of CA does except SF. That's why SF is a criminal paradise.
Blueprint: Not a late April Fools joke.
For months, @ManoRajuPD has defied a court order to take new cases, playing politics with public safety.
Result? Defendants go without lawyers and some are even released.
Enough is enough.
https://www.sfblueprint.org/take-action/tell-public-defender-mano-raju-do-your-job

people are sleeping on how excellent goose has become under the hood (interface needs some work but team is pushing).
it's a superpower. https://github.com/block/goose

I don’t know if you guys have fully internalized that America is about to simultaneously invent superintelligence and monopolize outer space.

lots of guys saying they haven’t played a video game since @openclaw dropped
agents are the new dopamine fix

Most interesting thing is as he said himself that it's not an AI/tech biz that is $1B…

Karpathy's setup keeps a 400K-word research knowledge-base without RAG for LLM query.
Dump sources into raw/.
Let an LLM turn them into linked Markdown.
Let it add summaries, concepts, and backlinks.
View it in Obsidian.
Ask the wiki questions with an LLM.
Let it make notes, slides, or charts.
Feed those outputs back into the wiki.
Run checks for gaps and errors.
Andrej Karpathy: LLM Knowledge Bases
Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating

The product roadmap is getting replaced by direct customer input in the AI era.
@jack on what changes when the product IS the conversation: "When your interface is a conversation with your customer, instead of this visual navigation, you suddenly get this amazing fidelity of, like, what do our customers actually care about?"

has anyone created gstack/compound-engineering for authors?

My biggest takeaways from @simonw:
1. November 2025 was an inflection point for AI coding. GPT 5.1 and Claude Opus 4.5 crossed a threshold where coding agents went from “mostly works” to “almost always does what you want it to do.” Software engineers who tinkered over the holidays realized the technology had become genuinely reliable.
2. Mid-career engineers are the most vulnerable—not juniors, not seniors. AI amplifies experienced engineers by letting them leverage decades of pattern recognition. It also dramatically helps new engineers onboard. Cloudflare and Shopify each hired a thousand interns because AI cut ramp-up time from a month to a week. But mid-career engineers who haven’t accumulated deep expertise and have already captured the beginner boost are in the most precarious position.
3. AI exhaustion is real and underestimated. Simon runs four coding agents in parallel and is mentally wiped out by 11 a.m. He’s getting more time back, but his brain is exhausted from the intensity of directing multiple autonomous workers. Some engineers are losing sleep to keep agents running. This may just be a novelty issue, but the underlying dynamic—that managing AI amplifies cognitive load even as it reduces labor—is a real tension. Good companies will manage expectations rather than expecting 5x output indefinitely.
4. Code is cheap now. This simple idea has profound implications. The thing that used to take most of the time—writing code—now takes the least. The bottleneck has shifted to everything else: deciding what to build, proving ideas work, getting user feedback. Since prototyping is nearly free, Simon often builds three versions of every feature when he’s getting started.
5. The “dark factory” is the most radical experiment in AI-assisted development happening right now. A company called StrongDM established a policy: nobody writes code, nobody reads code. Instead, they run a swarm of AI-simulated end users 24/7—thousands of fake employees making requests like “give me access to Jira”—at $10,000 a day in token costs. They even had coding agents build simulated versions of Slack, Jira, and Okta from API documentation so they could test without rate limits.
6. "Red/green TDD" is the single highest-leverage agentic engineering pattern. Having coding agents write tests first, watch them fail, then write the implementation, then watch them pass produces materially better results. The five-word prompt “use red/green TDD” encodes this entire workflow because the agents recognize the jargon.
7. “Hoarding things you know how to do” is one of Simon's other favorite agentic engineering patterns. Simon maintains a GitHub repo of 193 small HTML/JavaScript tools and a separate research repo of coding-agent experiments. Each one captures a technique, a proof of concept, or a library he’s tested. When a new problem arrives, he can point Claude Code at past projects and say “combine these two approaches.”
8. The "lethal trifecta" makes AI agent security fundamentally unsolved. Whenever an AI agent has access to private data, exposure to untrusted content (like incoming emails), and the ability to send data externally (like replying to email), you have a lethal trifecta. Prompt injection—where malicious instructions in untrusted text override the agent’s intended behavior—cannot be reliably prevented. Simon has predicted a “Challenger disaster” for AI security every six months for three years. It hasn’t happened yet, but he’s pretty sure it will.
9. Start every project from a thin template, not a long instructions file. Coding agents are phenomenally good at matching existing patterns. A single test file with your preferred indentation and style is more effective than paragraphs of written instructions. Simon starts every project with a template containing one test (literally testing that 1 + 1 = 2) laid out in his preferred style. The agent picks it up and follows the convention across the entire codebase. This is cheaper and more reliable than maintaining elaborate prompt files.
10. The pelican-on-a-bicycle benchmark accidentally became a real AI benchmark. Simon created it as a joke to mock numeric benchmarks—get each LLM to generate an SVG of a pelican riding a bicycle, and compare the drawings. Unexpectedly, there’s a strong correlation between how good the drawing is and how good the model is at everything else. Nobody can explain why. It’s become a meme: Gemini 3.1’s launch video featured a pelican riding a bicycle. The AI labs are aware of it and quietly competing on it.
Don't miss our full conversation: https://www.youtube.com/watch?v=wc8FBhQtdsA
Lenny Rachitsky: "Using coding agents well is taking every inch of my 25 years of experience as a software engineer."
Simon Willison (@simonw) is one of the most prolific independent software engineers and most trusted voices on how AI is changing the craft of building software. He co-created

Really enjoyed editing @karrisaarinen’s Thesis essay over on @every today. Think Asimov’s laws of robotics, but for AI agents https://every.to/thesis/how-to-design-for-human-agent-interaction

This is very true in my experience
Programming used to feel kind of leisurely and relaxing, like a crossword or sudoku
Now it feels more like debate. My raw horsepower is at the limit of my ability to absorb complex information and make decisions with real consequences.
Lenny Rachitsky: "Using coding agents well is taking every inch of my 25 years of experience as a software engineer, and it is mentally exhausting.
I can fire up four agents in parallel and have them work on four different problems, and by 11am I am wiped out for the day.
There is a limit on

Staying power matters in careers just as it does for startups
If in the right seat, better to stay there and compound then keep jumping around

Tip: Figure out your personal ceiling for running multiple agents in parallel.
We need to accept that more agents running doesn't mean more of _you_ available. The narrative is still mostly about throughput and parallelism, but almost nobody's talking about what it actually costs the human in the loop.
You're holding multiple problem contexts in your head at once, making judgment calls continuously, and absorbing the anxiety of not knowing what any one agent might be quietly getting wrong.
That's a new kind of cognitive labor we don't have good language for yet.
I've started treating long agentic sessions the way I'd treat deep focus work: time-boxed and tighter scopes per agent dramatically change how much mental overhead each thread carries.
Finding your personal ceiling with these tools is itself a skill and most of us are going to learn it the hard way before we learn it intentionally.
Lenny Rachitsky: "Using coding agents well is taking every inch of my 25 years of experience as a software engineer, and it is mentally exhausting.
I can fire up four agents in parallel and have them work on four different problems, and by 11am I am wiped out for the day.
There is a limit on

Marc Andreessen calls OpenClaw "one of the 10 most important software breakthroughs"
Bold claim, but I completely agree.
I've been setting up founders with OpenClaw and watching solo operators multiply their output in real-time.
Tasks that ate up their time every day/week are just melting off their todo list and getting owned by OpenClaw "employees."
The wildest part is how fast it compounds. Start with one small workflow, and a month later they're delegating things they never thought they could.
Latent.Space: 🆕 Marc Andreessen’s 2026 AI Thesis: Agents, Open Source, and Why This Time Is Different
https://www.latent.space/p/pmarca
@pmarca of @a16z says AI people keep swinging between utopian and apocalyptic for one simple reason: this field has been “almost here” for 80 years. But now, the

As San Francisco supervisor, Connie Chan famously said property crimes like theft and burglary weren’t “serious crimes” and did not warrant effort from SFPD, which she was advocating defunding at the time.
💥Susan Dyer Reynolds🗞️: Connie Chan running for Pelosi’s seat lying through her teeth:
“Criminals should go to jail.”
The real Connie Chan:

woke up and my mentions are full of these
Both me and @davemorin tried to talk sense into Anthropic, best we managed was delaying this for a week.
Funny how timings match up, first they copy some popular features into their closed harness, then they lock out open source.
mvpr: @bigben7182000 @steipete

Anthropic just sent an email saying that you can no longer run 3rd party harnesses like OpenClaw using Claude subscriptions.
Right now, both OpenAI and Anthropic are losing money on power users who run multiple agents 24/7 using their $100-200 subscription plans.
This reminds me of when Uber and Lyft were subsidizing rides to win market share. After both companies went public in 2019, ride prices nearly doubled over the next few years. And it took Uber 14 years from its founding before the company posted its first profitable year in 2024.
Both OpenAI and Anthropic are likely to go public soon. Once this happens, their margins will be public as well and there will be a lot of scrutiny on the money-losing all-you-can-eat subscriptions.
So I think there's a good chance that these subscriptions will either get more expensive or more limited after both companies go public.
The counter-argument is that both companies might keep prices low as long as there's still heavy competition and that compute costs will drop as well.
But yeah, overall I don't think the unlimited buffet will last forever.
Running local models on Mac Minis and Mac Studios is looking more appealing now as a safety net.
Peter Yang: Nooooo

I, for one, admire that San Francisco progressive blog @MLNow was willing to nuke the political career of their favorite supervisor @JackieFielder_ for a story. I was not familiar with their game.
Lee Edwards: So this story is getting crazier.
City says @MLNow has a page for page copy of this confidential memo.
Fielder’s office denies being the source of the leak.
If Chan also denies it, then one of them is straight up lying.
To the press, public, City Attorney. 🫠
More…