Ali Rohde

One job that’s only going to grow in the AI era? Personal security for AI leaders. Sad that's where we are.

Maybe. I personally get frustrated that knowledge workers default to Claude Cowork because it sounds like the product “for them,” when Claude Code is more powerful and just as easy to use if you can weather the brief adjustment period. I appreciate that OAI has one app for everyone, but you’re probably right that “Codex” makes non-engineers assume it’s not for them

Codex is a gazillion times better than Claude Code or Cowork for knowledge work. Needs a rebrand desperately.

@apoorvjain25 @ycombinator @LightconePod @koomen agree the teams set up is so much harder, probably needs additional infrastructure to be used at scale

Worth saying out loud: everyone in AI right now is building individual AI. Personal copilot, personal memory, personal context. Team AI is a fundamentally different problem. The moment two people share a brain, you need permission-aware retrieval, source conflict resolution, and citations tied to evidence per user. The hardest property to build, the one that gates everything else, is ACL fidelity at retrieval time. Without it, the shared brain leaks. Pete's "self-improving dream cycle" framing nails the team layer specifically. Spent the last year building Pulse on the same thesis. Shared org brain, MCP-native (lives inside Claude / Cursor / Codex), ACLs mirrored from each source system at retrieval, never expanded. Every sentence cited or marked inferred. YC built theirs for YC. We built ours for any team. Important conversation to have in public.

Over the past year, we've been building our own internal agent infrastructure at YC: over 350 tools, self-improving skill loops, and a shared organizational brain that gets smarter overnight. In this episode of the @LightconePod, we sat down with YC General Partner Pete @koomen to talk about how he led the effort from the ground up. We cover how giving agents unrestricted access to one database was the key unlock, the self-improving skill loops that get smarter overnight, and why he thinks we've arrived at the personal computer moment for AI. 00:39 — YC's AI Stack 02:15 — The Finance Team Problem That Started It All 05:07 — SQL Access Changes Everything 07:20 — One Database to Rule Them All 09:14 — Jevons Paradox 10:07 — Denormalizing for Agents 12:15 — The Single-Player Era of Agents 14:16 — 350 Tools and a Shared Registry 16:24 — Skillify, DRY, and MECE Resolvers 18:23 — The Self-Improving Dream Cycle 20:26 — The Two-Sentence Pitch Skill 23:06 — How Super Intelligence Compounds 25:10 — Recording Everything as a Building Layer 27:10 — The Shared Organizational Brain 29:18 — Trust-Default Culture as a Requirement 30:44 — Raising the Floor for New Employees 32:35 — Horseless Carriages 34:24 — Why Chat Is the Best Interface for Agents 38:50 — Just-in-Time Software 40:49 — Centralizing vs. Decentralizing AI 43:32 — The Personal AI Revolution

@ycombinator @LightconePod @koomen I do appreciate the ensemble format, not that common these days, feels like the cast of Friends chatting at Central Perk

@thsottiaux Just discovered Codex App’s /side command and love it tiny docs FYI: it works in the app, but I don’t think it’s listed on the App commands page yet

Thinking back, quite crazy to think about the progress that was made since December last year when GPT-5.2 was released. It now is seeing <1% of usage in production and is considered quite outdated!

To simplify our Codex compute fleet management, we will be sunsetting GPT-5.2 and GPT-5.3-Codex in Codex on June 2nd when logged in with your ChatGPT account. For free plans, GPT-5.5 will be the default frontier model to build and work with going forward. These models will remain available on our API.

@TheTennisLetter goals

NOVAK DJOKOVIC HITS AN AROUND THE NET POST WINNER AT ROLAND GARROS. 39 years old and still doing things like this. 🤯🤯🤯🤯🤯

@netcapgirl in your manifesting era

one of the strangest phenomena in life is that nothing happens when you try really hard. achievement obv takes hard, sometimes grueling work, but the idea that you’re “trying” or “forcing” something to happen is the opposite of what actually causes it to happen

@tekbog Not sure companies need to create that many jobs to contribute meaningfully to GDP — suspect that traditional link will weaken (not go to zero, but weaken nonetheless)

a lot of jobs can and will be automated, but i think a lot of new markets will emerge, otherwise there’s no gdp and companies will just be nationalized and that’s it?

@tekbog @890691905140v lol

@890691905140v i think it already did

@Aiswarya_Sankar no better use case for AI then fixing problems caused by other AI

This is what we've been seeing with every company we work with. Try justifying spending 100k on token spend when only 18k even makes it to a stable prod feature. In the rush to maximize AI token spend, companies are wasting over 44% on bug fixes

Agree. This was a refreshingly honest and rare take, and 100% right for Uber. What have they shipped in the last 5 years that actually moved the needle? That’s not a diss: it’s an incredible marketplace business. Their biggest threat is self-driving, not adding more little features to the app. You could almost take it one step further: what are all those engineers doing there at all?

Uber’s COO has said that it’s getting “harder to justify” its AI costs because there was no way to show a link between AI spend and any meaningful increase in useful features. This is the first time I’ve seen a company say this directly. businessinsider.com/uber-coo-andre…

@hnshah Started w a bunch of MCPs but have actually ripped out most of them in favor of using companies APIs directly

Thinking about MCPs (connectors / apps for AI) a bunch this week. What are you favorite ones?

@ivanburazin meanwhile... x.com/TechCrunch/sta…

Feels very Salesforce late-2024. Agentforce launched at ~$2/conversation, basically trying to treat agents as new revenue stream. Market pushed back, and Salesforce shifted toward Headless 360 — entire platform accessible through APIs/MCP/CLI. If the moat is enterprise lock-in, adding friction is how you teach people to route around you (albeit very, very slowly — I think we in startup land underestimate this). Microsoft feels earlier in that cycle. They still have to go through the stages of grief.

TIL reinforcement learning companies are creating fake versions of MS Office and not paying Microsoft a dime because MSFT is being stubborn and forcing their licenses upon agents like humans. x.com/ivanburazin/st…

@patowc @cyb3rops +1. I think similar to what @johncrickett expressed. Unfortunately that downward path is just so easy and is the default path. Great engineers will get a huge lift but I suspect for everyone else things will get worse, and the average enterprise codebase will get much worse, too

@cyb3rops I think there are two different paths and contexts depending on who is the actor using AI:

I think AI coding hype follows roughly four stages: 1. Amazement You try it and can’t believe how much code it generates from a few prompts. 2. Expansion You start more and more projects because shipping suddenly feels cheap and fast. This is also the phase where people start convincing everyone around them: - coworkers - management - friends in other companies because nobody wants to “fall behind” in 6–12 months. That creates a massive snowball/FOMO effect. 3. The grind phase You realize the generated code has architectural issues, sloppy mistakes, weird abstractions, duplicated logic, broken edge cases, etc. So you start: - re-prompting - switching models - increasing reasoning effort - reviewing fixes - generating fixes for previous fixes And suddenly you spend your days reviewing AI-generated pull requests instead of building software. 4. Realization You realize AI coding increases output much faster than it increases certainty. The code still needs: - review - testing - ownership - architectural understanding - long-term maintenance Usually by expensive senior engineers. And the interesting thing is: this whole cycle can take many months or even more than a year because people become socially and professionally invested in the narrative themselves. Once teams, managers, and entire companies have been convinced that this is the future, it becomes psychologically and politically very hard to later say: “Actually, the ROI is much lower than we expected.”

@johncrickett @cyb3rops +1

I think 3 is a mistake. All these are fiddling with knobs, not engineering: - re-prompting - switching models - increasing reasoning effort Instead, the lesson here is to apply good engineering practices. Ensure the context is clear. Ensure the requirements are clear. Have clear acceptance tests. Leverage automated checks (compilers, linters, test-automation). Review the code. Work in small increments (which makes it easier to review the code).

@romainhuet @reach_vb @Andy_AJT 🔥🔥🔥

The energy around Codex in Europe right now is incredible! I met founders, enterprises, and developers adopting Codex in all kinds of ways. The London meetup was one of my favorite moments: nearly 200 builders joined last minute to hang out and build. Such a special community!

Loving Codex. @thsottiaux small UX thing I ran into this morning: I had a weirdly hard time finding a shortcut to close or archive a thread in Codex Desktop. I expected Command+W to close the current thread and drop me back into New Chat, like Claude Code does on macOS. Eventually found that Command+Shift+A archives the current thread, but it seems pretty hard to discover. I still couldn’t find a non-archiving close shortcut. Might be worth documenting Command+Shift+A somewhere, and I’d still love Command+W to close a thread view without archiving it.

Codex is our WorkGPT

@rpenacastro thank you sir. figured there was something just hadn't looked it up yet

@RohdeAli No no no, you should be doing this with Claude Code using the DOCX native skill for formatting. This is literally me seeing you scold the AI and wasting tokens:

@staysaasy this is very good

> Hello 100x engineer, you’ve spent $100k in tokens this month. What have you to show for it > I was building a harness for my AI tooling setup. Nothing that impacts the company bottom line. > Sounds good to me. FYI we’re going to go layoff half the company because we’re over budget. Keep up the good work buddy.