Some examples of how to make it dramatically safer to use coding agents. And how to prevent your company from throwing the coding AGI baby out with the bathwater.
1️⃣ Preview & Branch Deployments
When your agents pushes code, Vercel gives it a URL they can visit with an exact replica of your infra, running in a production-like environment. With our db integrations like @neon, this is a full stack replica, with serverless staging Postgres DBs that can branch too.
This also allows code review and QA agents to focus on product functionality, rather than just doing static scans of code that point out stuff that might be inconsequential in the grand scheme of things.
2️⃣ Instant Rollbacks & Atomicity
Each deployment defines its entire universe. The CDN configuration. The Routing code. It partitions caches. As LLMs would say, this doesn’t just make rollbacks instant, it makes them reliable.
Within 300ms, we rollback the “pointer” to a set of infra resources. Same unique ability that powers Skew Protection and Rolling Releases. Ofc this is great for human error too 😁
3️⃣ First-class Flags support
Agents can now invoke 𝚟𝚌 𝚏𝚕𝚊𝚐𝚜 and start writing Flags-native code thanks to our Skill. Any company that's built products at scale, like Meta and Google, use flags as a key de-risking mechanism. Flags allow you to progressively release product capabilities, as well as roll them back. They allow cohorts of users including, crucially, its creators, to test the product in prod.
Scaling the organizational best practice and discipline to use flags is tricky for humans. We now make it natural and easy for agents.
4️⃣ Anti-fragile Compute
If a single-node server runs out of memory or crashes, at best you drop traffic, at worst you have to manually repair it. Our compute layer can tolerate a lot of human and AI mistakes. If a function crashes, the next request goes to a healthy one.
And unlike traditional serverless, in the event of a backend timing out or a provider becoming unhealthy, we dynamically adjust concurrency and only bill for Active CPU.
5️⃣ Designed for Autonomy
Autonomy is the new Observability. Instead of staring at charts and wiring up alerts, we automatically detect anomalies in error rates, traffic, and usage. Vercel Agent can react (i.e.: ) and autonomously investigate and find root causes. The full Vercel platform can be automated via CLI, MCP and APIs, which helps you automate any workflow or repair via coding agents.