Chai Landau

If you work in tech you've probably come across the "taste" discourse. One morning Silicon Valley woke up and decided they were done with copying each other's landing pages and that typefaces were suddenly important. Everyone is pointing to moodboards of vintage Mercedes SLs, The Row flip flops and yohji yamamoto fits and saying "look, I have good taste." Tech CEOs in Vineyard Vines polos have thrown the word around so much it's lost all meaning. People have argued its importance, tried desperately to define it and, even worse, argued why their taste is better than everyone else's. The biggest problem with the taste discourse is it misses what makes creative work good or interesting. Taste implies a consensus for what counts as a good creative choice vs a bad one, decided by gatekeepers. Anyone working in the creative scene knows you can have excellent taste and still produce work that's completely uninteresting. Long before the taste discourse existed you had creatives lifting their identities from the Instagram explore page to cosplay as someone with taste, pulling from the same pool of references as everyone else. I've been in plenty of rooms (gallery parties, store openings, screenings) full of stylish, tasteful and ultimately boring people. The word we're looking for is character. Character is the intangible quality of someone with a completely original perspective who expresses it in their work and identity. Someone who violates narratives, subverts norms and makes stuff that divides opinion. When you see a memorable brand identity, you're seeing the accumulation of unique experiences and references from a weirdo. Advertising has always celebrated its eccentrics and given them a seat in the boardroom. Tech is catching up to the value of individuality, but it's overcorrected into an infatuation with a word nobody can define. Reggie James put it best: "The problem with most tech taste is they don't dare scare the hoes nearly enough." (Meme by VC Brags)

Better Auth is joining Vercel. Bereket Engida and his team built the auth library that developers have made their own, with 4.6M+ weekly npm downloads and more than 850 contributors. Welcome, Bereket and team! https://lnkd.in/gmt4xBDz

Last week was a huge week at Speakeasy... We had two new hires start on the same day, our first ever onboarding class! First, David Danialy is joining us as a Product Engineer working on our AI Control Plane. David has come out of the gates absolutely firing, in just his first week he built support for personal account tracking into the dashboard, which is critically important for every business trying to get a handle on AI usage. Second, Sarah C. is joining our growth team helping our customers map out their AI needs and getting them started on the Speakeasy platform. She's bringing boundless energy to the SF office, reinforcing that I am officially old, and absolutely delighting all the prospects and customers she works with. If you want my next hiring post to be about you, check out our careers page. We have a number of open roles!

Today we're excited to share how we are evolving the Koah brand! https://lnkd.in/ggQGy73g Huge shoutout to Chris Reilly, Isabelle Rossi de Leon, Katie D., Mike Choi, and the whole Koah team for the work that went into this brand and our latest products.

June was a crazy month for Speakeasy. The platform is getting deeper by the day. We started with agent security, now we're helping with how agents connect, what they spend, and what they're allowed to do. This are some of the highlights from the past few weeks: 🔌 MCP Private Tunnels. Your internal servers, the ones you'd never expose publicly, can now be governed on Speakeasy without a public endpoint in sight. 💸 Cost Attribution. See cost by team, agent, user, session, even by MCP server, scoped all the way down to the action. 🧠 Prompt-Based Policies. Some risks aren't a regex problem, they're a judgment call. Now an LLM judge makes that call for you, in plain language you write yourself. 🛂 Enterprise Managed Auth. See every OAuth connection in your org in one place and shut any of them down instantly. July is going to be even bigger. Stay tuned !

We built a platform called Block App Kit that lets everyone at Block safely build and deploy the internal tools that they need when they need them, removing the bottleneck on engineering resources. Below is a TL;DR of what worked, what we learned and what we're still figuring out after shipping Block App Kit to the whole company earlier this year. The full writeup is available on the Block engineering blog (and is linked in the comments). What worked was clearly separating platform responsibilities from agent responsibilities along deterministic vs non-deterministic lines. The platform is responsible for everything that makes an app safe to deploy and use (identity, access control, secrets, the deploy path, etc.) so that the agent cannot make an unsafe choice. The agent is solely responsible for coding the app on top of the platform. This means that a finance analyst with no engineering background can describe the tool she wants in plain language and watch an agent build it on her laptop in minutes, then deploy it behind SSO without making a single unsafe decision along the way. What we learned is that shipping, not development, is the bottleneck now. Getting something working on localhost is easy, but a vibe-coded app isn't really a tool until other people can safely use it. Inside a company that handles real financial data, that is a high bar and security and data integrity needs to be a first-class feature, not an afterthought. Solving that problem by building a platform that all employees can safely build on and trust turned out to be the bulk of the work. What we're still figuring out is the ideal human interface. Today, roughly four in five builders sit outside of engineering, across more than fifty orgs, with a catalog well over a thousand apps. Generating an app is the easy part now, so the harder questions are the human ones. How much should the agent decide on its own, and where does a builder need to stay in the loop? When does the platform need to step in and ask someone to check the work (e.g. that the data query powering a dashboard is correct)? The shape of that interaction does not have a clear precedent yet, and it is the next problem that we're tackling with Block App Kit. Huge shoutout to my amazing teammates Brad S., Drew Crutchfield, Lucas Isaza, Morgan Martin, Nathan Thillairajah, Phil Azar and a big thanks to our leads Bradley Axen and Erika Rice Scherpelz for all the support along the way. I've had so much fun building with you all and I feel so grateful to be part of this team. P.S. We're looking for a Principal Engineer to join our team, link in the comments!

We built a platform called Block App Kit that lets everyone at Block safely build and deploy the internal tools that they need when they need them, removing the bottleneck on engineering resources. Below is a TL;DR of what worked, what we learned and what we're still figuring out after shipping Block App Kit to the whole company earlier this year. The full writeup is available on the Block engineering blog (and is linked in the comments). What worked was clearly separating platform responsibilities from agent responsibilities along deterministic vs non-deterministic lines. The platform is responsible for everything that makes an app safe to deploy and use (identity, access control, secrets, the deploy path, etc.) so that the agent cannot make an unsafe choice. The agent is solely responsible for coding the app on top of the platform. This means that a finance analyst with no engineering background can describe the tool she wants in plain language and watch an agent build it on her laptop in minutes, then deploy it behind SSO without making a single unsafe decision along the way. What we learned is that shipping, not development, is the bottleneck now. Getting something working on localhost is easy, but a vibe-coded app isn't really a tool until other people can safely use it. Inside a company that handles real financial data, that is a high bar and security and data integrity needs to be a first-class feature, not an afterthought. Solving that problem by building a platform that all employees can safely build on and trust turned out to be the bulk of the work. What we're still figuring out is the ideal human interface. Today, roughly four in five builders sit outside of engineering, across more than fifty orgs, with a catalog well over a thousand apps. Generating an app is the easy part now, so the harder questions are the human ones. How much should the agent decide on its own, and where does a builder need to stay in the loop? When does the platform need to step in and ask someone to check the work (e.g. that the data query powering a dashboard is correct)? The shape of that interaction does not have a clear precedent yet, and it is the next problem that we're tackling with Block App Kit. Huge shoutout to my amazing teammates Brad S., Drew Crutchfield, Lucas Isaza, Morgan Martin, Nathan Thillairajah, Phil Azar and a big thanks to our leads Bradley Axen and Erika Rice Scherpelz for all the support along the way. I've had so much fun building with you all and I feel so grateful to be part of this team. P.S. We're looking for a Principal Engineer to join our team, link in the comments!

Got to sit down with Rob Alderson at one of my new favourite publications, Undo, to talk AI, creativity, and how I've been finding my way into film. He clocked me as "inspired, but anxious" 🙃 which, tbh, is probably fair. I don't hide my concerns about where this is all headed, even as I stay curious and endlessly inspired with what it opens up. "It's... complicated" as Rob puts it. Undo is an excellent publication with a lot of journalistic integrity, so please do consider supporting and reading the other articles — they're well worth your time. :) https://lnkd.in/gkf4c2Fz