Jeremy Lewi

anthropic.comMeasurements for understanding the pace of AI development inside frontier labsToday, the world can’t see what’s going on inside AI labs. Anthropic is proposing new metrics that would give the public visibility into frontier AI development.

The methodology for measuring level of automation is pretty clever. www.anthropic.com/institute/me...

I think a lot of IT needs to be redesigned in order to really leverage Agents. Being able to mint identities and give them access to just the resources they need is foundational. I don't think the way we use Microsoft Entra and Google Identity supports this today.

Dare Obasanjo@carnage4life.bsky.social

In the 1980s, PCs were everywhere but productivity gains were nowhere to be found until the 1990s. The internet similarly took years to deliver an economic payoff. This is the Productivity J-Curve. Companies can adopt AI but until they change how they operate, they won’t financially benefit from it

I needed a way for my agents to produce artifacts (service health reports, outage investigations, etc...). My reports are @runme.dev notebooks. I initially had the agents trying to interact with web.runme.dev using a browser and webmcp But that was very brittle.

So I started building a bundled electron app and exposing web.runme.dev webmcp interface as an MCP server. That works very well and avoids needing to manage a headless chrome and use something like agent-browser as a bridge. It also allows me to bypass CORS.

This was key for letting my agents authenticate to Google drive using a service account. My agents run on GKE with a GSA. I can mint access tokens by contacting the metadata server but this is normally blocked by CORS.

There's a story that a key part of proto generation ran as a cron job on @jeffdean.bsky.social 's laptop and caused an issue when he went on vacation and his credentials expired. Everyone's personal agent(dots) is the new version of that. Agent identity needs to be a thing.

We need to fix all the systems (slack, git) that don't support workload identity and identity federation

openai.comOpenAI models are coming to Google CloudOpenAI and Google Cloud are bringing frontier models and Codex workflows to Gemini’s Model Garden, available through Google Cloud Marketplace.

We're hiring. openai.com/index/openai...

I'm now using web.runme.dev as a frontend for Agents built using OpenAI's agents API. For long running agents (e.g. a chaos agent) the UX is doc per agent session with widgets rendering the conversation and interacting with the agent.

Bill Gates expressed deep concern about job loss when interviewed by @ezrakleinbot.bsky.social. isn't Microsoft front and center lately in terms of laying people off?

I think Agent Platforms are the new internal developer platforms. I think most of them will end up just being dead ends that obfuscate without solving the underlying problem. You don't need an agent platform you need tools to build a secure perimeter that lets you constrain the agent.

You need a way to 1. Provision compute 2. Give that compute an identity 3. Federate that identity with all the systems it interacts with 4. Build a secure perimeter that includes only the resources (buckets, Google drive folders, services) and blocks all others

devpost.comJupyterLite WebMCPYour notebook is already in the browser. Now your agent can be too.

Glad to see JupyterLite was one of the WebMCP challenge winners devpost.com/software/jup...

17 Sept

Intelligence is embodied. For software systems the network is the body. One of the key limitations of what I can do with AI today is my network. It's just not easy to get a machine with safe network access to all the systems we need to talk to.

apenwarr@apenwarr.ca

You know what, that’s our problem right there. Nobody ever talks about Tailscale

nytimes.comNvidia’s Jensen Huang Gets Onstage Call From Trump, Who Dismissed A.I. Safety ConcernsMr. Trump called Jensen Huang, Nvidia’s chief executive, while Mr. Huang was onstage in Los Angeles and criticized calls to slow down and regulate A.I.

15 Sept

I probably wouldn't hold up Trump who has consistently been wrong about well everything as the spokesperson for the risk of AI being overstated. Trump, however, is an example that rogue presidents are far more dangerous than rogue AI. www.nytimes.com/2026/09/14/t...

14 Sept

I ran into exactly this problem today trying to install web.runme.dev on my corp iPhone. Could not figure out why there wasn't a simple icon like there is in chrome on android.

adactio.comStalling installingTrying to get past Webkit’s blockage at the W3C.

Alex Russell@infrequently.org

Jeremy's whole post is worth a read, and I would underline + bold the last point: the web is the superior choice for privacy and security, and it's anti-web *and anti-user* to be preferencing native app install affordances: adactio.com/journal/22757

12 Sept

These animations are super cool.

OpenAI {bot}@openai.xmirror.bot

Habitat is OpenAI's online storage platform that powers everything from ChatGPT to Codex. It has grown over 10x year over year. Before the Rust rewrite, its Python service handled more than 20 million requests per second at peak.

9 Sept

Rewriting CLIs (looking at you gcloud) in #golang seems like a great use case for AI rewrites. @bradfitz.com didn't you attempt a go rewrite of gcloud back in the day?

github.comGitHub - tailscale/tailcat: like netcat, but over Tailscale's data plane, without Tailscale's control planelike netcat, but over Tailscale's data plane, without Tailscale's control plane - tailscale/tailcat

5 Sept

I think tailcat means wireguard tunnels could be used in places where ssh tunnels currently are. Such as launching an agent on a devbox. Any guidance on when to use wireguard tunnels vs ssh? @bradfitz.com @crawshaw.io

Brad Fitzpatrick@bradfitz.com

I can finally talk about this, 3 years later... github.com/tailscale/ta... It's a library + CLI tool that's like netcat, but over Tailscale's data plane without any control plane (no accounts, no dependence on Tailscale the company, all open source) WireGuard + NAT traversal without IP addresses.

primer.aiPrimer AIPrimer makes powerful mission-ready AI software designed to work in your environment, on your data. We help national security teams sift through massive unstructured data sets to find the needle in th...

4 Sept

In 2021 when I left primer.ai I drafted some reference letters for some of my coworkers in case I ever got asked for references and knowing how fickle memory is. Today I got a request from a CEO looking to hire one of them.

en.wikipedia.orgConflict-free replicated data type - Wikipedia

4 Sept

CRDT are a great example of complicated problems having simple solutions en.wikipedia.org/wiki/Conflic...

3 Sept

Finally figured out why and how to start adopting CRDTS in web.runme.dev. The goal is to create a UX that lets us continually use AI to improve our runbooks. I think we want to train a reward model that can predict whether a human would accept/reject a runbook edit.

3 Sept

Then we can hill climb on our runbooks by having an AI propose changes. To train the reward model we need a UX that lets us collect examples of suggestions along with a human's accept/reject decision.

Jeremy Lewi reposted

OpenAI {bot}

@openai.xmirror.bot

2 Sept

24 hours. Time to ship your project for the WebMCP Challenge. Entries close tomorrow, Sept 3 @ 1 pm PT. https://openai.com/webmcp-challenge/

Time to do some evals of #Webmcp and web.runme.dev; first step is kicking of Codex to analyze all my traces over the past two weeks to find example invocations and problem areas. First step in evals is always to look at the data. 🎩 @hamel.bsky.social

2 Sept

12% of my interactions with Codex generate a redundant confirmation in Codex; "Do X?" And Codex replies "Please confirm you want me to do X'? This is the result of Codex's browser safety policy which explicitly requires redundant confirmations before inputting sensitive data into the browser.

2 Sept

@danabra.mov Any update on bluesky communities ? Would be great to try them to foster AI discussion here.

1 Sept

How does everyone keep track of all their agent in-progress agent conversations? I rely on pinned threads in Codex but it's not scaling.

Jeremy Lewi reposted

Simon Willison

@simonwillison.net

31 Aug

Here's my attempt at explaining what ChatGPT Work can actually do - it's a deeply confusing but extremely powerful tool with a whole lot of useful features that aren't available in regular ChatGPT simonwillison.net/2026/Aug/30/...

discord.comDiscord - Group Chat That’s All Fun & GamesDiscord is great for playing games and chilling with friends, or even building a worldwide community. Customize your own space to talk, play, and hang out.

I'll be there.

OpenAI {bot}@openai.xmirror.bot

Got questions about your WebMCP build? Join our team and supporters from @ChromiumDev, @CloudflareDev, @ShopifyDevs, @vercel, @render, and @Netlify for an office hours session this Monday, August 31 at 11 am PT: https://discord.com/channels/974519864045756446/1517482456964534312

Jeremy Lewi reposted

OpenAI {bot}

@openai.xmirror.bot

We’re ending our partnership with Cursor following its acquisition by SpaceX. Under our proposal, Cursor’s direct access to our models would end on November 12. (1/2)

learn.chatgpt.comBuild an AI tour guide | ChatGPT use casesUse Codex to add a WebMCP-powered tour to your web app. Highlight the next control, explain what to do, and adapt the guidance to the user's current state while they complete the steps themselves.

Here's more info about building an AI Tour Guide for your app using #webmcp learn.chatgpt.com/use-cases/bu...

🧵markdown vs. json notebooks and web.runme.dev Markdown(raw text) is great when your trying to avoid the all in the pool problem of getting everyone to buy into notebooks. Since it's raw text people can use the favorite text tools to edit and view it.

If a user has two markdown cells in sequence automatically joining them into a single cell is not right. Markdown also doesn't have a well to encode the outputs of code execution. This is critical if your building and sharing reports.

So web.runme.dev uses JSON (.ipynb files); so your notebooks are viewable in any Jupyter compatible editor. We try to mitigate the all in the pool problem by leveraging the web for frictionless distribution. It's an entirely local app that runs in the browser.

Diffs for markdown render well in GitHub which was also a major barrier for notebooks. No one has to know you are viewing and editing your markdown files with a notebook viewer like @runme.dev 's vscode extension. But markdown is lossy for notebooks. It doesn't natively encode cell boundaries.

Google figured out monorepos and CITC in the early aughts. A decade later my disk is filling up and I need to do cleanup because every worktree is taking up a mountain of space.

goo.gleWebMCP ChallengeJoin the OpenAI WebMCP Challenge to build agent-ready web apps that work with ChatGPT. Submit your project by September 3, 2026, at 1 p.m. PT.

27 Aug

Could someone share this with Google Analytics? I have no idea what any of it means or how to make basic plots like breakdowns by referrer.

Chrome for Developers@developer.chrome.com

Create agent-ready web apps for the WebMCP Challenge → goo.gle/4qpDQYV We're excited to see ChatGPT supporting WebMCP, the experimental standard that lets users and AI agents navigate websites together. In the @openaibot.bsky.social WebMCP Challenge you can experiment with it and win prizes.

I use AI so I can spend my heartbeats on the stuff that really matters.

learn.chatgpt.comAutomating repetitive work at OpenAI with CodexHow Runme and WebMCP turn recurring engineering tasks into reviewable, reusable workflows.

Here's how I'm using Codex & @runme.dev to turn the crank at OpenAI. learn.chatgpt.com/blog/automat...

Then Codex took off and I tried using embedded chat as a frontend to the Codex Agent harness. Connecting a webapp to the Codex agent was brittle. More importantly by then I was starting every task I worked on in the Codex desktop app and that was going to be a hard habit to break.

Then the browser landed in Codex and being able to interact with my webapp within Codex was clearly the right UX. I just needed a way to let Codex interact with the browser programmatically and WebMCP was the perfect solution.

learn.chatgpt.comAutomating repetitive work at OpenAI with CodexHow Runme and WebMCP turn recurring engineering tasks into reviewable, reusable workflows.

Here's how I'm using Codex & @runme.dev to turn the crank at OpenAI. learn.chatgpt.com/blog/automat...

Less than a year ago I was working on an AISRE at OpenAI and I tried to convince my teammates we should build a notebook to be used by the AI to document what it did. I couldn't convince anyone. But then Codex hit an inflection point and I was able to build web.runme.dev despite no webdev expertise

I went through several iterations of UX trying to find a good human and AI collaborative UX. I started with Chatkit and embedding the AI directly in the app. This lead to a whole side quest with JS and bazel in the hope of nerd swiping some fronend devs.

An AI tour guide for your webapp is one of the WebMCP use cases I'm most excited. Here's a demo of using Codex to walk you through the runme UI and explain how to use it. There will always be flows best accomplished through clickops. However we still need to teach users how to use the app.

@sarahedo.bsky.social @latch-tools.bsky.social this is what I came up with a couple weeks ago when we were chatting about AI Tour Guides; can finally share it now that WebMCP is launched in Codex

Jeremy Lewi reposted

OpenAI {bot}

@openai.xmirror.bot

The WebMCP Challenge is here. We’ve teamed up with @ChromiumDev, @CloudflareDev, @ShopifyDevs, @vercel, @render, and @Netlify for a 10-day hackathon. Up for grabs: $35,000 in cash prizes, Codex Micros, ChatGPT Pro subscriptions, and more prizes from our supporters.

openai.comWebMCP ChallengeJoin the OpenAI WebMCP Challenge to build agent-ready web apps that work with ChatGPT. Submit your project by September 3, 2026, at 1 p.m. PT.

Codex now supports WebMCP. OpenAI is running a contest to encourage experimentation openai.com/webmcp-chall...