Shawn Wang
Related people
Episodes
- How Anthropic ships the #1 LLM in the World safely: Separating the Brain from the Hands
- Why Agents Love Supabase
- 🔬 Synthesis Superintelligence: from Semiconductors to Superconductors — Periodic Labs’ Liam Fedus and Ekin Dogus Cubuk
- Academia is for Ambition — Alex Zhang, MIT
- Building Legal Agents When You Can't See the Data - Jacob Lauritzen, CTO of Legora
- Dyna-2: General Purpose Physical Agents with Hour-Long Autonomy — Jason Ma, Dyna Robotics
- AI for the Fortune 2000: How Tenex Builds an AI-Native SDLC — Alex Lieberman & CJ Hess
- The Programmable Data Center — Evan Conrad, SF Compute
- ⚡️ The 2026 Software Factory Standard — Zach Lloyd, Warp.dev
- The Recommendation Engine for Commodity Traders — Anekha Sokhal, Moshi
Notes
Latent Space: The AI Engineer Podcast · ⚡️ The 2026 Software Factory Standard — Zach Lloyd, Warp.dev
-
Why Does Warp Stay Out of Token Reselling?
Zach Lloyd says paying by the API is 5 to 50 times more expensive than buying a pile of subscriptions. Small companies he talks with want the factory but stay on a rotating collection of $200 subscriptions, and that rotation is the cheapest way. They are probably not running anything like 17 different subscriptions. He does not know whether the terms allow the rotation. The other benefits disappear, it does not scale, and past a certain company size it will not be allowed.
Shawn Wang adds that a $200 plan includes $8,000 of API usage, and Lloyd agrees that is not money to ignore. Lloyd does not want Warp in the pure coding-agent business, because a competitive coding agent cannot be built o
Latent Space: The AI Engineer Podcast · Why Agents Love Supabase
-
How Much Does Supabase Spend on AI Coding?
Paul Copplestone says Supabase does not set one AI coding policy for the whole company. Product teams already choose their own tools and habits, and coding agents were left the same way. A few cost rules are starting to line up. He still wants each team to decide.
Shawn Wang asks how much the company spends and whether the bill is a worry. Copplestone answers with who decides, not with a company-wide cap. Because the company is remote, decisions are written down, so he can ask an assistant what the bill is. The figure he gives is about $300,000 a month for about 450 people. He first says 600 and corrects it to 450. Hiring is fast, so the bill moves with headcount when the tools give the tea
-
Why Were AI Coding Agents Breaking Postgres?
Paul Copplestone says a database is harder to leave than a frontend tool, which is why Supabase can stay stickier than a coding app such as Lovable. Shawn Wang had asked whether a product chief ends up more attached to the coding app or to the database. Lovable was also the first of the coding apps to stop its agent from doing damage.
Early agents stuffed image data into Postgres and turned off row-level security just to make a project run. Lovable wrote skills that blocked the moves. He treats the restraint as the useful part. Newer models have stopped making the same mistakes on their own, so the failure is less common than it was.
-
Why Do AI Coding Agents Keep Choosing Supabase?
Shawn Wang says his tracking of model answers shows frontier models recommending Supabase as the default integrated backend: the database, auth, functions, and the other APIs in one place. He traces the lead to being in the market for three or four years before the models took off, already the open-source alternative people compared with Firebase.
Paul Copplestone calls the timing an accident and says the job now is to keep the lead. The early habit of spending time with the community left posts, videos, and podcasts about how to build on Supabase, and open source made that material easy for a model to read.
-
Why Do AI Coding Agents Activate Faster Than Human Signups?
Paul Copplestone says the latest jump in Supabase signups came from the models getting better, not from a marketing spend. Most of what they spend on attention is still memes.
A flood of lighter traffic usually hurts activation and the share of people who start paying. Both went up. The agent finishes signup and activation in one step, so activation is almost 100 percent. The new users also started paying at a higher rate than developers who signed up the old way.
Shawn Wang pulls up the signup chart and says the curve keeps forcing a new axis. The unusual part, in Copplestone's account, is that activation and payment rose together as the audience got broader. Friction at the front door wa
-
Where Did the Supabase Co-founders Meet?
Shawn Wang says Paul Copplestone and Ant Wilson, the Supabase co-founders, went through Entrepreneur First in Singapore, each on a different company, and stayed after the cohort as housemates.
Paul Copplestone says seven people lived in one house, each running a different startup, before Supabase. An accelerator usually pairs a technical founder with a domain expert. Everyone in the house was a builder. Pairing two builders got more done on developer tools, because a developer is already the subject-matter expert. The co-founders kept building together from there, and they hired most of the other people who had lived in the house. Rory, one of them, went on to run go-to-market.
Latent Space: The AI Engineer Podcast · How Anthropic ships the #1 LLM in the World safely: Separating the…
-
Why Does Anthropic Separate an AI Agent's Brain From Its Hands?
Katelyn Lesse says an agent's hands should only reach the network places the setup allows. Untrusted code can run in the sandbox, and egress can be an allow list or shut off entirely.
Claude Managed Agents ship that architecture out of the box. As the platform team learns a new control, customers get it without rebuilding the harness. She groups the new vocabulary as brain and hands, credentials, injection, and egress. Nine months earlier, nobody was using those words. They are new problems, and the team wants customers to pick up the fixes while Anthropic is still solving them in its own infrastructure. Internally, the same systems are being moved to secure by default.
She agrees that clo
Latent Space: The AI Engineer Podcast · The Recommendation Engine for Commodity Traders — Anekha Sokhal…
-
Why Is Moshi Hosting Its First Webinar?
Anekha Sokhal says Moshi is hosting its first webinar because inbound demand is too heavy to cover with one-to-one meetings alone. After an initial meeting, the team wants a clear picture of the problem and a plan, then a way to stream the rest of the demand.
She had meant to leave New York the same day. Meetings then landed on Thursday, Friday, and Monday, plus energy meetings in Houston, filling a full week after the conference. Her team kept flagging another signup. On the buy side, banks set the meetings up and brought people into small rooms for a round table. A CFO doing investor relations would meet five funds at once, and she was one of the five funds.
Shawn Wang has not got on-sit
Latent Space: The AI Engineer Podcast · AI for the Fortune 2000: How Tenex Builds an AI-Native SDLC — Alex…
-
Why Won't Large Companies Rip Out Salesforce and SAP?
Alex Lieberman says ripping a system out and rebuilding it gets harder as a company moves up market, and he does not agree that Salesforce or SAP is really at risk. Those products stay because so much money is at stake. Getting the replacement wrong can mean hundreds of millions or billions of dollars, so the company stays risk averse.
For a mid-tier tool that costs a few million dollars a year, the question is whether that spend will ever move the needle enough to put engineers on a rebuild. Tenex built its own CRM and its own operating system, then ran into the other half of what software is worth: the maintenance and improvement you do not have to think about when the tool is not the cor
-
How Much Do Companies Spend on AI Tokens Per Employee?
Shawn Wang says most companies spend about $10 to $200 per employee on AI tokens. AI-native companies turn the comparison into a percentage, roughly 10%. An average AI startup employee is about 1x, so token spend matches employee spend, and the range runs up to 5 to 10x token spend.
On raw tokens, Tenex lands between one to one and two to one, and the ratio comes off as pretty high. The question on the team is when to reach for the biggest model. Costs have lately been driven way down from Opus 5.5. For a stretch, everyone used Astra and Fable because the results were the best, but the models are expensive. Token maxing and token leaderboards flipped into too much cost, and the overrun was
-
What Does Paying Engineers for Outcomes Mean?
CJ Hess says paying engineers for outcomes means rewarding output instead of hours, because hourly client work paid him less as his output got better and faster. He liked the ethos Alex Lieberman first posted on Twitter: reward engineers for their output as opposed to rewarding them for their time.
Fast forwarding two years, engineers can parallelize a ton of work and have much more output at a time with Sonnet 3.5 and Gemini 2.5 Pro. He took the bet on much higher output and a higher earning capacity from his skill set and a deeper lean into AI, though the upside might not have been obvious. He fell in love with the model. It is what the future of engineering looks like. When AI is solving
Latent Space: The AI Engineer Podcast · 🔬 Synthesis Superintelligence: from Semiconductors to…
-
How Does Periodic Labs Work With Semiconductor Companies?
Liam Fedus says Periodic Labs works with semiconductor companies by sending its own engineers and researchers on site, into very private and secure labs, to integrate the company's AI and computation with the partner's process. The lab has been its own first customer, building the tools for its own materials discovery and materials engineering. It is now taking that know-how into other industries, with a huge focus on semiconductors.
Those forward-deployed engineers run inference locally and can also train on the partner's data, so the partner owns the intelligence built from that data. Fedus says that beats handing over an API or an untrained model and leaving the partner to figure it out.
Get Together · Instant Pot CEO Robert Wang on how his small pressure cooker…
-
Who Is Robert Wang
He is the founder of Instant Pot. More than 5.5 million people follow Instant Pot related accounts on Facebook alone. People even make decals and knit sweaters for their Instant Pot.
-
What Was Wang’s Opinion of His Product
He thought the problem wasn’t that it was going to be popular on the market, but that stores had limited coverage. The best way to let people know of his product was through online businesses.