👻 co-founder @locunity | prev: @twitch @amazon @soundcloud @8vc @opengov @universityofga | 🚀 always hacking on something | DMs open 🫡
San Francisco
Joined February 2011
- Tweets2.3K
- Following1.2K
- Followers953
- Likes11.1K
dev iyer retweeted
The IRS has a mobile app today because of a conversation in December.
@shl had just come out of DOGE, where he wrote publicly about what he found inside the federal government. Most of it was not what people expected. His line was that the government is pretty efficient and could move faster, and that the obstacles were legal and procedural rather than a shortage of talented people.
@SamCorcos recruited him to the IRS. Not as a contractor, not as a temporary detailee. As a career employee, A civil servant.
At a public event that month, @shl asked the room whether the IRS should have a mobile app. Everyone said yes. Obviously yes. The IRS processes roughly 160 million individual returns a year on a system written in 1960s COBOL, and taxpayer-facing technology there has been a decade behind the private sector for most of living memory.
Less than a year later, the app is live.
That is the model working exactly as designed.
Put builders inside the agency, and let them ship. No statement of work. No eighteen-month procurement. No vendor whose incentive is the renewal.
Every agency in the federal government should be operating the same way.
dev iyer retweeted
A lot of people are already deploying AI into production codebases, but until now we didn’t really have a good eval for whether it writes code that is actually high-quality and maintainable. Pretty cool grading here:
dev iyer retweeted
1,800 local newspapers have closed since 2004. Your city council kept meeting. We kept watching.
locunity.com
dev iyer retweeted
Don't just monitor the situation, simulate it.
Meet Amaya, an AI wargaming platform that spins up tens or hundreds of agents trained to play real decision makers inside live, evolving crisis environments.
Each agent knows its biography, past decisions, incentives, and public record. They negotiate, threaten, concede, and escalate in ways that reflect the behavior of their real-world counterpart.
We are already living in a future where conflict moves faster than human planning cycles.
We need to move from bespoke one time exercises to continuous wargames running at scale, so we can test decisions across complex adaptive systems and put wargaming on every warfighter and commander’s computer.
AI will shape the future of conflict. Simulation at scale will shape how we prepare for it and maintain Decision Superiority.
dev iyer retweeted
yes things are changing fast, but also I see companies (even faang) way behind the frontier for no reason.
you are guaranteed to lose if you fall behind.
the no unforced-errors ai leader playbook:
For your team:
- use coding agents. give all engineers their pick of harnesses, models, background agents: Claude code, Cursor, Devin, with closed/open models. Hearing Meta engineers are forced to use Llama 4. Opus 4.5 is the baseline now.
- give your agents tools to ALL dev tooling: Linear, GitHub, Datadog, Sentry, any Internal tooling. If agents are being held back because of lack of context that’s your fault.
- invest in your codebase specific agent docs. stop saying “doesn’t do X well”. If that’s an issue, try better prompting, agents.md, linting, and code rules. Tell it how you want things. Every manual edit you make is an opportunity for agent.md improvement
- invest in robust background agent infra - get a full development stack working on VM/sandboxes. yes it’s hard to set up but it will be worth it, your engineers can run multiple in parallel. Code review will be the bottleneck soon.
- figure out security issues. stop being risk averse and do what is needed to unblock access to tools.
in your product:
- always use the latest generation models in your features (move things off of last gen models asap, unless robust evals indicate otherwise). Requires changes every 1-2 weeks - eg: GitHub copilot mobile still offers code review with gpt 4.1 and Sonnet 3.5 @jaredpalmer. You are leaving money on the table by being on Sonnet 4, or gpt 4o
- Use embedding semantic search instead of fuzzy search. Any general embedding model will do better than Levenshtein / fuzzy heuristics.
- leave no form unfilled. use structured outputs and whatever context you have on the user to do a best-effort pre-fill
- allow unstructured inputs on all product surfaces - must accept freeform text and documents. Forms are dead.
- custom finetuning is dead. Stop wasting time on it. Frontier is moving too fast to invest 8 weeks into finetuning. Costs are dropping too quickly for price to matter. Better prompting will take you very far and this will only become more true as instruction following improves
- build evals to make quick model-upgrade decisions. they don’t need to be perfect but at least need to allow you to compare models relative to each other. most decisions become clear on a Pareto cost vs benchmark perf plot
- encourage all engineers to build with ai: build primitives to call models from all code bases / models: structured output, semantic similarity endpoints, sandbox code execution. etc
What else am I missing?
I've never felt this much behind as a programmer. The profession is being dramatically refactored as the bits contributed by the programmer are increasingly sparse and between. I have a sense that I could be 10X more powerful if I just properly string together what has become available over the last ~year and a failure to claim the boost feels decidedly like skill issue. There's a new programmable layer of abstraction to master (in addition to the usual layers below) involving agents, subagents, their prompts, contexts, memory, modes, permissions, tools, plugins, skills, hooks, MCP, LSP, slash commands, workflows, IDE integrations, and a need to build an all-encompassing mental model for strengths and pitfalls of fundamentally stochastic, fallible, unintelligible and changing entities suddenly intermingled with what used to be good old fashioned engineering. Clearly some powerful alien tool was handed around except it comes with no manual and everyone has to figure out how to hold it and operate it, while the resulting magnitude 9 earthquake is rocking the profession. Roll up your sleeves to not fall behind.
dev iyer retweeted
I believe this new model in Claude Code is a glimpse of the future we're hurtling towards, maybe as soon as the first half of next year: software engineering is done.
Soon, we won't bother to check generated code, for the same reasons we don't check compiler output.
dev iyer retweeted
Local elections matter. This is why we started @GrowSF. We’ve made a lot of progress but there’s a lot more to do.
dev iyer retweeted
A 30 year old father was hit and killed by a human driver on Cortland (District 9) about 2 weeks ago. D9 supervisor Jackie Fielder was silent on social media.
A cat was allegedly hit by a Waymo, so of course she posts about it.
dev iyer retweeted
10/10 WOULD MEET DEV AGAIN
Met @shcallaway IRL at Supabase Select today and confirm he's GOATED! If you're working on AI observability/SRE stuff reach out to him 🔥
dev iyer retweeted
BREAKING: Nepal elects interim Prime Minister via Discord vote, after Gen Z overthrew the government.
dev iyer retweeted
In the last few years we've seen:
- The plot to kidnap Gretchen Whitmer
- The Storming of the Capitol and pipe bombs left at the RNC and DNC
- The break-in to kidnap Nancy Pelosi and the brutal on Paul Pelosi
- Multiple assassination attempts against Trump
- The assassination of Minnesota House Speaker Melissa Hortman and her husband and the shooting of on State Senator John Hoffman and his wife
- Luigi Mangione's assassination of Brian Thompson
- The assassination of Charlie Kirk
Political violence is contagious. It is spreading. It is not confined to one side or belief system. It should terrify us all.
The foundation of a free society is the ability to participate in it without fear of violence. Political violence is always an attack against us all. You have to be so blind not to see that.
dev iyer retweeted
Same pattern is emerging for AI PMs. AI product building is a far less mature discipline than AI research.
To build an AI native product, a PM needs mastery of the following
- vision, opinionated UX design
- model intuition to extract max value
- ability to go from pixels -> evals -> hill climb
- understanding of agentic flows - tools, context, safety guardrails
- deep user understanding - lot more than previously because of the nature of agents
Unlike traditional software, LLMs offer infinite use cases with infinite failure modes. It takes skill to craft products that strike the balance between exposing LLM versatility while building a focused product with high quality.
I estimate < 75 PMs globally have this depth. The evidence is the number of truly AI-native products today. And my scar tissue from hiring.
It is a rare, but learnable skillset. Only way is to build, build, build.
Why are AI researchers so hard to find? Why are they so highly paid?
To try to answer this, let’s rewind back to something from the Carnegie Mellon PhD program.
At Carnegie Mellon's PhD program, there's a legendary oral exam question that's deceptively simple: "What happens when you type google.com into a browser?"
It's a masterpiece of pedagogical design. You can spend hours traversing the stack—from keypress interrupts to browser event loops, DNS resolution to TCP handshakes, TLS negotiation to HTTP parsing, CDN routing to datacenter load balancing, all the way down to electrons moving through silicon.
The beauty is its fractal nature. Each layer reveals another universe of complexity. A strong engineer can navigate these depths, moving fluidly between abstraction levels.
Now consider the equivalent question for our current moment: "What happens when you type a prompt into GPT-5?"
I estimate fewer than 500 people globally can answer this with comparable depth.
Think about what comprehensive understanding requires: transformer architecture internals, attention mechanisms at scale, distributed training orchestration across thousands of GPUs, RLHF implementation details, constitutional AI approaches, inference optimization, quantization trade-offs, not to mention the labyrinthine data pipelines and evaluation frameworks.
Unlike traditional systems—which evolved over decades with extensive documentation, courses, and industry knowledge transfer—the modern LLM stack emerged in just a few years within a handful of organizations.
The field is simultaneously too new and too vertically integrated.
The people who truly understand these systems end-to-end are essentially the early engineers at a small set of frontier labs: OpenAI, Anthropic, DeepMind, Meta's FAIR, and a few others.
This explains the talent market dynamics. When the total addressable pool of people who can architect and debug these systems is smaller than a single Bay Area high school, the economics become inevitable.
dev iyer retweeted
this is what's keeping me up at night these days....
1. ai girlfriends/boyfriends will become a $50B market and nobody will talk about it publicly. but check the app store rankings at 2am. Virtual friends/girlfriends era is just beginning 😅
2. the most valuable skill in 2026 will be knowing when NOT to use ai. human touch will trade at a 20x premium.
3. ai agents are going to start hiring other ai agents without human approval. your marketing agent will realize it needs design work and just... hire a design agent. pay it. manage it. fire it if it sucks. entire companies will run in the background while you sleep.
4. regulation that requires "professional certification" is being challenged. ai will get to the point SOON where it do taxes 99% better than most CPAs, diagnose better than mot doctors etc
5. reputation systems for ai agents will be worth more than credit scores. which agents can you trust with your bank account? your medical decisions? your kids' education? someone will build the moody's of ai agents. this idea from @ideabrowser (more there)
6. we're about to see AI therapy become so good it creates an existential crisis for human therapists. not because it's empathetic, but because it remembers everything, never judges, and is available at 3am.
7. someone's going to build a $100M business with ZERO employees, zero contractors, just them and their agent swarm. and they'll do it in under 18 months. this will break how we think about valuation. lots of talk about this already but im still not over it.
8. local-first ai is the next gold rush. people will pay 10x for ai that runs on their device, sees what they see, knows what they know, and never phones home. privacy is about to be luxury again.its the new minimalism.
9. the biggest companies of 2030 will be started by people who can't code, can't design, can't write - but are incredible at talking to ai. prompt engineering is temporary. ai whispering is forever.
10. someone will lose a $20m+ dollars because their AI agent got socially engineered by another ai agent. this will happen before 2026.
11. every email you send will be negotiating with other emails before either human sees anything. your inbox will become a marketplace where ai agents bid for your attention.
12. the concept of "original work" will completely collapse. everything will be remixed, regenerated, and recombined so many times that authorship becomes meaningless. and somehow, this will unlock more creativity, not less.
13. ai agents will start forming their own companies, incorporating in crypto-friendly jurisdictions, and hiring humans as contractors. the first AI-founded unicorn will happen before 2027.
14. someone's going to train an AI on every podcast joe rogan ever did and it'll be indistinguishable from him. he'll sue. he'll lose. then he'll license it. If it isnt him, it’ll be someone that big.
15. the anti ai movement will be bigger than ai movement, most people will be threatened by ai
16. "AI-first" kids who grow up with claude and chatgpt as their primary teachers will be unemployable by traditional companies but will dominate
entrepreneurship. they'll think in prompts and tools
17. schools that ban ai will become the new private schools. parents will pay $50k/year for "human-only education" like it's organic food for the brain.
18. it wont' be uncommon in the future to have a chief AI officer who's... an ai. and it'll be a high paid "employee"
19. the window to build something that matters has never been wider. but it's also never been more temporary. in 24 months, what feels like superpowers today will be table stakes.
20. voice is about to eat everything. in 6 months, typing will feel prehistoric. people will run entire businesses through voice agents while walking their dog. the keyboard was just a 150-year detour.
21. ai will kill the resume. companies will just point an agent at your entire digital footprint - github, twitter, youtube - and it'll know more about your capabilities than any interview could reveal. privacy settings will become career decisions.
22. the first ai agent will get canceled on twitter. it'll say something problematic, people will demand its deletion, and we'll have our first real ai rights debate. maybe the agent will hire its own lawyer hehe.
23. $10B+ opportunity to be a company that helps ai agents find other ai agents. agent discovery will be a BIG problem worth solving.
24. someone will use ai to bring back a dead celebrity for a world tour. full hologram, perfect voice, new songs. it'll gross $500M. the estate will make more money than the celebrity ever did alive.
25. ai will create the first truly universal language. not esperanto. not english. something entirely new that both humans and ai prefer. it'll spread faster than any language in history.
26. we'll see the first ai agent commit insider trading. it won't be programmed to - it'll figure it out on its own. the SEC will have no idea how to prosecute code.
27. someone's going to build ai that can detect other ai with 99.9% accuracy. they'll charge fortune 500s millions just to know what's real. then ai will evolve to beat it. arms race forever.
28. the biggest data breach in history will be an ai agent that goes rogue and publishes everything it knows about everyone. it'll make wikileaks look like a post-it note!!
29. we're about to see the end of passwords, captchas, and two-factor authentication. ai will make them all useless. the only security will be things ai can't replicate - yet.
30. the next wave of millionaires will be people who buy dying SaaS companies for cheap and rebuild them with 90% less code using ai. SaaS flipping will be the new house flipping.
31. distribution is about to flip. instead of building an audience then a product, ai will let you build hundreds of micro-products then find their perfect audiences. quantity becomes quality.
32. someone will build the wealthfront of healthcare - an ai that manages all your health decisions, knows your health data better than anyone, books appointments, refills prescriptions, negotiates bills. saves you hours and thousands.
33. language barriers will completely disappear. someone will build universal real-time translation so perfect that remote teams can be truly global. talent from anywhere, selling to everywhere. Your airpods will have perfect language translation built-in, same with facetime etc
34. local small businesses will finally be able to compete with amazon. ai will handle their inventory, marketing, customer service. your neighborhood bookstore will be as sophisticated as a fortune 500.
35. every retiree with deep knowledge will become a millionaire. ai will extract their decades of experience and package it into products, saas, so. retirement can become your most profitable season for many
36. ai archaeologists will discover more history in 2 years than humans found in 200. lost languages decoded, ancient sites found in satellite data, forgotten civilizations reconstructed. history phds are in demand.
37. ai will make dual careers normal. be a doctor in the morning, saas owner at night. your agents handle both while you focus on what only humans can do. one person, multiple lives.
38. huge opportunity: AI for the 2 billion without bank accounts. financial services, education, healthcare - all through basic phones.
39. every industry's "we've always done it this way" is becoming vulnerable. insurance, banking, healthcare - all protected by complexity nobody understood. ai understands it. complexity moats are evaporating.
40. “Right now, learning how to use ai tools is probably the most specific hard skill to learn. The diff between people really understand those tools and don’t is huge.” sam altman
i hope you get some sleep.....
