Nov 1, 2025 · 42m · latent-space
⚡️ Ship AI recap: Agents, Workflows, and Python — w/ Vercel CTO Malte Ubl
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of Latent Space, Vercel CTO Malte Ubl breaks down the key announcements from Vercel's Ship AI event, covering the Workflow Development Kit, AI SDK v6, autonomous DevOps agents, Python runtime support on Fluid Compute, and Vercel's dogfooding-driven engineering philosophy.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
The hosts as informed peer 6.3 Guest teaching 3.6 Guest disagreement 1.4 The hosts pushing back 2.7
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 14:21 Malte reframes model lab vs framework developer mindset
Malte firmly counters the host's framing, arguing that model research labs do not care about practical developer constraints like 30-second streaming latency.
Hardest push from the hosts ▶ 13:22 Swyx pushes back using frontier lab architectural philosophiesSwyx explicitly states he will push back, arguing Vercel's bottom-up abstraction approach clashes with frontier labs' model-centric Hollywood principle.
Biggest teaching moment ▶ 23:33 Malte explains LLMs do not do raw anomaly detectionMalte corrects Swyx's assumption about LLMs doing time-series analysis by explaining that an independent time-series pipeline triggers the agent, which then inspects text logs.
The host holds their own ▶ 4:48 Swyx explains workflow execution and suspension mechanicsSwyx demonstrates deep domain expertise from his time at Temporal, detailing infinite execution loops and pause/resume serverless cost models.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Introducing Vercel's Workflow Development Kit | 8 | 2 | 1 | 2 | Swyx draws directly on his deep background at Temporal to break down the mechanics of durable workflows, serverless pause/resume models, and webhook-based human-in-the-loop approvals. Malte affirms the shared history and expands on Vercel's zero-cost pause model and open-source strategy. | |
| Workflow Directives and AI SDK Evolution | 6 | 4 | 1 | 2 | Swyx probes Malte on the controversial syntax of React/Vercel directives and the competitive landscape of AI SDK against Mastra and LangChain. Malte outlines the architectural reasoning for keeping AI SDK low-level and unopinionated. | |
| Framework Design Principles and Dogfooding | 7 | 4 | 3 | 6 | Swyx openly challenges Malte by contrasting Vercel's bottom-up web developer tooling against frontier labs' model-first 'Hollywood principle' architecture. Malte rejects the dichotomy, pointing out that model labs don't build consumer-facing applications where streaming UX is critical. | |
| Vercel Agent and AI DevOps Observability | 6 | 5 | 2 | 4 | Swyx expresses skepticism regarding AI SRE reliability and the difficulty of language models parsing time-series anomalies. Malte clarifies Vercel's architectural division: time-series anomaly detection runs in a dedicated database pipeline, while the agent handles query execution and log inspection. | |
| Internal Custom Agents and Forward Deployment | 5 | 3 | 1 | 1 | Malte walks through internal custom agents built at Vercel for sales lead qualification, abuse prevention, and data warehouse queries, alongside their forward-deployed engineering engagement model. Swyx connects these initiatives to dogfooding at enterprise scale. | |
| Python Support and Fluid Compute Infrastructure | 6 | 3 | 1 | 2 | Swyx asks about Vercel's historical 'bet on JavaScript' stance versus the new first-class Python support and compares fluid compute overhead to workflow steps. Malte explains the low-level VM infrastructure enabling multi-runtime zero-config execution. | |
| AI Engineering Leadership and Safe Vibe Coding | 6 | 4 | 1 | 2 | Swyx inquires about organizational structures and safe vibe-coding boundaries for non-technical contributors, comparing proposed solutions to WorkOS. Malte reveals Vercel's work on extracting authentication and data access layers to protect applications against incompetent human or AI developers. |
Statements from this episode (17)
Ubl: Transaction systems require explicit or ad hoc workflow abstractions
“What's extremely common and essentially every transaction processing system that has ever created is that there either is an explicit abstraction for workflows in it, or someone made one ad hoc, because otherwise the thing just doesn't work.”
Vercel paused workflows incur zero compute cost indefinitely
“No, it literally does not cost anything. So yeah, you can run compute for infinite amount of time. You can then also, whenever, like, one of these steps fails, automatically retry them.”
Ubl: Vercel leads the truly open-source deployment business model
“Vercel maybe has not invented this, but certainly kind of is the most successful at a model where you say, okay, I have this software library and it's truly open source. Everyone can run it. It comes with like adapters for every place on the planet, and that m…”
Vercel AI SDK v6 stabilizes direct agent abstractions
“What it does introduce as a stable feature, because it's already as kind of experimental in, in AISDK five is a direct agent abstraction, which so far wasn't there, right? People would build agents with AISDK, but they would have to do it in Whitmore Baybones …”
Thick abstractions in nascent AI frameworks are usually wrong
“And so if you put a very thick abstraction, then it's probably going to be the wrong abstraction. You have to, so you have to be humble and say, okay, I need to stay low level. So that this can be flexibly used as trends emerge, right?”
Frontier AI model labs ignore application-level integration needs
“Like the model apps couldn't care less because they're not really building applications.”
Vercel never releases developer abstractions without internal dogfooding
“The absolute founding principle of a cell is that we never give you an abstraction that we haven't used ourselves. Like dog footing is ultimately the thing, right?”
Vercel AI SDK was extracted from v0 before re-integrating it
“Like that, like AISDK was extracted from B-Zero and then we build it and we kind of diverged a little bit and then we took on the substantial work to bring back VZero actually fully hosted on ISDK”
Framework creators build ivory towers unless they build real applications
“The reality is that framework builders are usually not application builders. And so they built ivory towers that when they're hyper geniuses, or they get lucky, they're happy to be good. But like, but if you want to do this in a reproducible fashion with a hig…”
Vercel DevOps agent will soon access codebases to open fix PRs
“One thing that we don't do today, but we'll do in the future is that we also give the Model X that particular agent access to your source code so that it can, first of all, figure out like, what does the error message mean? Right. And every time you can actual…”
AI agents only succeed when scoped to narrow problem domains
“At this moment, they're, like, agents are both extraordinarily effective and still very ineffective. And you have to find the right problems. And then when you find the right problems, they are super magical. And if you wander beyond, then they don't work.”
Asking employees what they hate identifies high-impact AI agent tasks
“Basically where the idea is that you go around your company and you ask people like, what do you hate most about your job? And I really think it finds the sweet spot because it finds problems that are, they're boring because they're tedious and repetitive, but…”
Vercel enterprise contracts require clients to build three AI agents
“We basically sign contracts with companies saying, okay, you have to commit to building three agents, and if you do, we're going to help you, like, we're going to build the first one for you. And then the second one, we are going to be there essentially by you…”
Vercel shipped zero-config deployment for Flask and FastAPI
“What we have done over the last weeks is we have shipped like zero config support on Vercel for all the popular Python framework like Flask and FastAPI.”
Vercel will eventually support PHP and Ruby runtimes
“We will eventually do PHP and Ruby, but we also, I think, care a lot about the details, and so the Essentially the reason why we haven't done it yet is because we do invest substantial amount of time to make the DX actually good and feel native to the ecosyste…”
AI product pivots must naturally extend a company's core identity
“And I think there's a lesson there to be learned for companies that haven't done the transition yet. Is that you have to do something that feels native to your company. And so the two big bets that we made early on, one being vZero and one being AISDK, I think…”
Future app platforms must extract auth and authorization completely
“Auth cannot be part of the app, because they're not going to get that right, right? So, Auth has to be extracted from the app. In fact, which data you can see, they also cannot be under control of the app, because again, you're going to get it wrong, right? So…”