Follow builders, not influencers.
TL;DR
Albert teased Anthropic’s Mythos Preview, Cat Wu juiced Claude Code’s CLI tricks, and Peter Steinberger patched CodexBar with 2 providers plus billing fixes. Levie said agents are eating knowledge work, while Nikunj Kothari preached retention over launch hype.
They say Claude Mythos Preview is now in the hands of launch partners through Project Glasswing, just two months after Claude Opus 4.6. The vibe is big: Anthropic’s Alex Albert calls Glasswing one of the most consequential moments he’s seen there, which suggests this isn’t just another model drop.
He says one of the biggest lessons from reviewing a bunch of calls and transcripts is that open-ended verification can chew through tokens without actually improving the output. He’s planning to write up how to do it more efficiently — a useful note for anyone building LLM workflows at Anthropic or elsewhere.
Cursor’s design bet is pure flow: fewer controls, more direct manipulation, and AI woven into the work instead of bolted on. The pitch is simple but strong — let people stay in the zone and make software without fighting the UI.
Run `/powerup` to surface the Claude Code CLI features the team actually uses. It’s a small but useful nudge toward making the tool feel less like a black box and more like a power-user workflow.
They argue that designers and engineers often do their best product thinking in the abstract — but once Figma or the IDE is open, they default to making the thing instead of framing the problem. The takeaway: more designers should move into PM roles, because they’re already wired to think in product terms.
He says the jump from chatbot to real agent is here: Box Agent can now take an RFP, hunt through source docs, run tools, and draft the response while you do something else. His bigger point is that frontier model gains are still compounding fast, and that’s about to push agentic workflows into legal, finance, consulting, sales, and beyond.
He says we’re back in LISPy territory: code is data and data is code again, except the weird modern twist is Markdown. He also points out a Claude Code quirk where bash can still append to files outside the workspace, which is a reminder that agent tooling is only as safe as its escape hatches.
Paper Review in Prism is meant to act like a careful technical reviewer, not a grammar checker: it checks math, derivations, notation, units, structure, and whether claims are actually supported. The output lands as an editable LaTeX review file inside the project, so authors can fix issues in the same workspace where they wrote the paper.
He says too many early founders are optimizing for views and fundraising instead of product and retention — and it shows up fast when you check changelogs and see weeks of nothing. His advice: spend that energy on customers who pay; the VCs will notice later.
CodexBar 0.20 is out with Perplexity and OpenCode Go support, plus a nicer account switcher that doesn’t force a re-login. He also fixed the Claude token/cost inflation bug and merged session usage into provider history, so the menu bar now tracks 16 providers without the accounting mess.
She introduced a Personalized Podcast skill that turns anything into a podcast with two AI hosts, publishes it as an RSS feed, and plays it in your normal podcast app. Her favorite use case is remixing meeting transcripts so the AIs “eavesdrop” and react — a neat example of her “content for one” thesis.
Harness engineering turns code into the product and the process
The Takeaway: Ryan Lopopolo’s core bet is that the fastest way to scale software is to make the agent the primary engineer and the harness the real product.
Key Insights
The Story
Ryan Lopopolo works on frontier product exploration at OpenAI Frontier, building enterprise agent deployments with governance and safety. His background spans Snowflake, Brex, Stripe, and Citadel, which explains the mix of operator discipline and AI maximalism. His team spent months building a greenfield app with “0% human code” and eventually “0% human review” in the hot path, using Codex to write product code, tests, CI, docs, dashboards, and even repo-management scripts.
The philosophy is blunt: if the model is the worker, the harness is the workplace. That meant retooling build systems until they finished in under a minute, adding observability so the agent could diagnose itself, and encoding standards into markdown, skills, and review agents. Ryan’s favorite move is to turn every lesson into durable text: “The models fundamentally crave text.” When a timeout bug appears, he doesn’t just patch it; he has Codex update reliability docs so the fix becomes policy.
His bigger point is contrarian but practical: software teams should stop optimizing for human legibility first. As models improve, they can propose abstractions, resolve merge conflicts, and close loops across the stack. The job of the engineer shifts upward—to define invariants, set boundaries, and let the agent chew through the rest.
Levie said agents won’t erase work, just push it up a layer; Yang argued they’ll shrink teams, not ambition. Garry Tan flagged an unpatched file leak in Claude’s coding env, while Kothari called Anthropic’s revenue ramp absurdly fast.
Rauch said v0 now builds physics, not just UI, while Karpathy noted GitHub Gists have weirdly good comments. Levie argued AI efficiency creates more work, not less, and Tan called open source’s golden age.
Karpathy pushed “your data, your files, your AI.” Levie argued context beat raw model IQ in enterprise AI. Garry Tan said GStack kept shipping security fixes fast, while No Priors spotlighted Periodic Labs’ bet on atoms, not just text.
Claude plugged into Microsoft 365 everywhere, Swyx said Devin one-shot blog-to-code, and Peter Steinberger called out GitHub’s API as still not built for agents. Aaron Levie hit the context wall, while Garry Tan shipped a DX review tool from his own stack.
Claude landed computer use on Windows, Karpathy argued LLMs should build your wiki, and Amjad Masad pushed Replit deeper into enterprise sales. Peter Yang said Cursor 3 got out of the agent’s way, while Peter Steinberger warned AI slop was flooding kernel security with real bugs.
Steinberger called plan mode training wheels, while Thariq gave Claude Code a mouse-friendly renderer and Cat Wu showed sessions jumping phone-to-laptop. Masad framed Replit as an OS for agents, Rauch said Vercel signups compounded fast, and Anthropic’s infra tweaks swung coding scores by 6 points.
Levie said AI productivity hit the enterprise risk wall, while Weil argued proofs got cleaner, not just better. Agarwal floated public source code as the new prod debugging, and Data Driven NYC claimed one founder could run a company if agents handled the layers below.
Karpathy warned unpinned deps can turn one hack into mass pwnage, while Rauch and Levie said agents still need human guardrails and redesigned workflows. Meanwhile Claude Code got enterprise auto mode, Replit added built-in monetization, and Swyx spotted “Sign in with ChatGPT” already live.
Andrej Karpathy highlighted how LLMs can argue any side, suggesting we use it as a feature. Guillermo Rauch finally shipped his dream text layout, bringing his vision to life. Meanwhile, Amjad Masad claimed AI is democratizing app building and elevating top engineers.
Andrej Karpathy suggested leveraging LLMs' ability to argue any side as a feature. Guillermo Rauch turned text layout dreams into reality with Vercel's latest feature. Meanwhile, Amjad Masad claimed AI is democratizing app building, liberating top engineers for bigger challenges.