Matthew Binshtok

software engineer @vercel ex-@palantirtech infrastructure, agents, reliability

[Matthew Binshtok @abeanstalker]

@MelkeyDev excited to cook more together 🤝

[Melkey @MelkeyDev]

big dog @abeanstalker cooked on this🔥

[Guillermo Rauch @rauchg]

When Opus 4.5 came out, it was a one-way door to a new way of engineering. Agents now do most of our coding.

Knowing the inherent flaws and over-confidence of LLMs, we sent a clear message to our teams. Vibing and mission-critical infrastructure don’t go together.

We’re sharing some of our early internal guidance in how we’re “agenting responsibly”, prioritizing security, durability, and availability at all times.
https://t.co/b36GiE76Ue

[Malte Ubl @cramforce]

Very timeline @abeanstalker shipped this very related blog post https://t.co/wUAQpXdomv

[Matthew Binshtok @abeanstalker]

@beh_zod @vercel born in our DMs 🫡

[Behzod @beh_zod]

"The litmus test is simple: would you be comfortable owning a production incident tied to this pull request?"

Loved this provocation from @abeanstalker when he shared it internally. Glad to see it on the @vercel blog.

[Matthew Binshtok @abeanstalker]

AI slop doesn't look like slop anymore. It’s a clean diff with green CI and a convincing PR description. It looks production-ready.

That's what makes it dangerous.

Agent responsibly.

https://t.co/x3rMjLgWM6

[Matthew Binshtok @abeanstalker]

@dillon_mulroy means a lot coming from you. appreciate it sir 🤝

[Matthew Binshtok @abeanstalker]

We ran a production failover of our control plane database last summer to make days like today uneventful.

Satisfying to see the rehearsal pay off.

Proud of the crew for executing under real pressure 🫡

[Malte Ubl @cramforce]

Earlier today our oncall team performed the failover we exercised last summer under true emergency conditions.

As designed, Vercel platform serving from our 20 regions was never impacted by control-plane issues and API services+dashboard recovered after the failover.

[Guillermo Rauch @rauchg]

All software will be generative and generated.
Adjust accordingly.

[Malte Ubl @cramforce]

Vercel no longer does code freezes. The great change batch is more risk than folks taking responsibility if they want to ship something on December 25

[AI SDK @aisdk]

AI SDK 6

Introducing agents, tool execution approval, full MCP support, tool calling with structured output, DevTools, reranking, standard JSON schema support, provider tools, image editing, and so much more.
https://t.co/VztDF4tXJ6

[Vercel @vercel]

We experienced a service disruption on Oct 20, 2025. We take full responsibility and apologize to our customers.

Read about the impact, timeline, and reliability improvements underway.
https://t.co/pl0cUK8f0n

[Bloomberg @business]

Vercel has raised $300 million at a $9.3 billion valuation, alongside a $300 million tender offer. https://t.co/OlR90jcTSB

[Matthew Binshtok @abeanstalker]

part of our job at @vercel is to prevent outages... so @matheusfrndes and I tried to start one on purpose

here's the deep dive on a live production failover of our control plane database

[Vercel @vercel]

We recently ran a successful live failover of our control plane database from Azure West US to East US 2.

Our goal was to validate that Vercel can survive a complete regional failure of its core database.

Here's the full breakdown:
https://t.co/5ng7adhZLc

[Vercel @vercel]

Vercel AI Gateway is now generally available.

• Access hundreds of models
• Zero markup on tokens (including BYOK)
• No provider accounts needed
• High rate limits
• Failover for high reliability
• Sub-20ms latency
• AI SDK and OpenAI-compatible

https://t.co/AIRDf0dnW6

[Matthew Binshtok @abeanstalker]

spent my day failing over @vercel’s control plane database from one coast to the other and back. a few tense hours with @matheusfrndes but huge win overall

[Malte Ubl @cramforce]

Earlier today Vercel performed a successful test of a failover of our control-plane database to its backup region and back.

We periodically perform such tests to ensure we’re prepared to responding to major cloud outages.

While we collected some learnings, there was no disruption to service. Additionally, Vercel’s serving of end-user traffic is backed by a globally replicated database completely resistant to any cloud region becoming unavailable.

[Matthew Binshtok @abeanstalker]

@dillon_mulroy hmm I wonder who this could be

[Matthew Binshtok @abeanstalker]

who needs a cold brew?

[Aryaman @aryamankhawow]

There is literally nothing scarier than seeing this in a Friday incident