Jay

CEO - Them AI

Host - Hope of AI | AI Generated Optimism

[Jay @jk_helloworld]

Dudes creating some random HTML pages with animations and claiming I’ve built Jarvis is the new trend on Instagram.

OMG.

[Jay @jk_helloworld]

Everyone tracks cloud spend.

Almost nobody tracks AI spend by team.

That’s about to become a very expensive problem.

[Jay @jk_helloworld]

@JunaidAckroyd Voice programming. 😌

[Jay @jk_helloworld]

@naval It already does. Including this tweet.

[Jay @jk_helloworld]

X is where you build.

LinkedIn is where you announce that you're building.

[Jay @jk_helloworld]

Now all I see is service companies turning into software companies, selling their services as a productized model.

[Jay @jk_helloworld]

https://t.co/ON6jPvJbIF

[Jay @jk_helloworld]

https://t.co/evp4zbFn0Y

[Jay @jk_helloworld]

Would love to connect with people working on consumer research and GTM.

[Jay @jk_helloworld]

@ThePeelPod Exact reason for why I am building Them AI.

https://t.co/9Xsrv4Xm0s

[Jay @jk_helloworld]

@DarrelFrater DM

[Jay @jk_helloworld]

Claude working for y’all?!

[Jay @jk_helloworld]

This is interesting, maybe they're trying to build a better model out of it, like Anthropic Mythos using the outcome?

[OpenAI @OpenAI]

We’re expanding OpenAI Daybreak to help democratize patching vulnerable software at machine speed:

- Codex Security plugin: find, validate, and fix vulnerabilities right inside Codex

- The full version of GPT-5.5-Cyber model: a great model for trusted defenders

- Cyber Partner Program: powering products built on top of our best cyber capabilities for leading security companies to secure the world's software

- Patch the Planet: working with maintainers to secure critical open source projects

https://t.co/hyIi6gQmkm

[Jay @jk_helloworld]

@nikitabier @catturd2 @RealJamesWoods Good. The "grow fast" playbook was overdue for a reset — turns out the only durable growth lever left is being worth following. Annoying for the bot farms, great for everyone actually building.

[Jay @jk_helloworld]

@paulg Building with agents forces this on you fast. You watch a model expose your wrong assumption in real time, dozens of times a day. Hard to stay precious about being right when the loop keeps showing the receipts.

[Jay @jk_helloworld]

@sama The talent war is really a loop-design war now. The people who can architect how these models actually get used are worth more than another point on a benchmark. Huge get.

[Jay @jk_helloworld]

This is the multi-model endgame: open weights catching the frontier, custom silicon making them nearly free to run. Orchestration beats allegiance. Route each call to whoever's fastest and cheapest that hour, not to a logo.

[Simon Willison @simonw]

Really looking forward to one of the super-fast custom silicon inference providers like @GroqInc or @cerebras getting GLM 5.2 running

Cerebras has GLM-4.7, Groq is still mostly Llama 3.x and gpt-oss

[Jay @jk_helloworld]

The quiet shift nobody's pricing in: "skill" is becoming a primitive inside the harness, not a mode the user picks. The job stopped being prompting the model — it's designing the loop that decides when to reach for which skill.

[Aravind Srinivas @AravSrinivas]

Perplexity Computer is an agent harness that just keeps delivering. Deep Research is now a native skill inside Computer (you don’t have to explicitly think of using it as a standalone mode anymore as long as you’re using Computer), furthering the state of the art significantly

[Jay @jk_helloworld]

3 years ago the skill was writing the perfect prompt.

Today it's designing the loop: what context the agent gets, when it stops, which model handles which step, how it checks its own work.

The job quietly went from "prompter" to "loop engineer." Nobody sent the memo.

[Jay @jk_helloworld]

Two talks the AI-building world is sharing this week:

→ Head of Canva AI: 90% of their engineers run a subagent harness daily (a main Sonnet agent delegating to Opus + Haiku) to build the #1 design tool.

→ Creator of Claude Code: 100% of his code written by Claude since Opus 4.5, automated by loops, dynamic workflows, and routines.

Same idea, different names.

Subagent harness. Loops. Dynamic workflows. Routines.

It's all one discipline: loop engineering. The shift from typing prompts by hand to designing the system that prompts, finds the work, gates it with a test, and records what happened, on its own.

The talks prove it works. I wrote down how to build one.

Worth more than 100 YouTube videos on agents.

Read below 👇

[Jay @jk_helloworld]

Most devs still drive their coding agent by hand: prompt, wait, read the diff, prompt again.

That's holding the wheel of a self-driving car.

The job is shifting from "prompter" to "loop designer" — you build the feedback loop, the agent runs it. Whoever learns this first ships 5x.