Sergey Burykin

Senior Software Engineer at Uber🧑‍💻 Ex Foundation Engineer at Jitsu (YC S20) 🚀 Traveler 30 Countries 🌎 Code, tech, startups ✍️ All opinions are mine 💭

[Insecure Agents Podcast @insecureagents]

Solving the Agent Identity Crisis: Sergey Burykin on Securing 1,000 Agents in Production at Uber

"Agent permissions should not extend user permissions. The final set should be the intersection between the user and the agent."

@xtreding, a Senior Software Engineer on the AI Security team at Uber, talked to us at @aiDotEngineer World's Fair about how @Uber took roughly 1,000 agents to production and how agent identity was a key part of that puzzle.

We get into:
> Why an agent should be scoped to the intersection of user and agent permissions, never just one
> The multiple layers of agent security: identity, authorization, runtime guardrails, observability
> Why static OAuth scopes break for non-deterministic agents and what dynamic scopes need to do
> Why coarse-grained APIs like Slack and Luma are a real blocker, and why sandboxing is also important

TIMESTAMPS

(00:50) The agent identity crisis explained
(01:40) Multi-hop delegation: when agent A calls agent B who calls a resource
(03:00) The core rule: authorize on the intersection of user and agent permissions
(04:20) Building the secure token exchange and MCP Gateway from scratch
(05:40) Why no off-the-shelf option fit Uber's Java, Go, Python stack
(07:00) Okta Cross-App Access and central, auditable identity
(09:00) AI security as a multilayer cake, and the persona guardrail
(10:00) Runtime guardrails vs blocking access after the fact
(12:30) What every team keeps rebuilding vs what should be customized
(13:00) Static OAuth scopes vs the dynamic scopes agents need
(14:10) Coarse-grained APIs: Luma can delete your event, Slack ships one token
(17:00) Exchanging your Uber identity for scoped third-party tokens
(20:30) Sandboxing is the next thing, and ephemeral credential injection
(23:00) 1,000 agents in production and the framework-agnostic challenge
(24:00) Building a big tent around agent identity and auth
(26:00) Distinguishing agent traffic from human traffic in production

[Sergey Burykin @xtreding]

How to say "I love coffee" without saying a word.

[Praveen Neppalli @praveenTweets]

Agentic software engineering adoption is on fire at @Uber. 1,800 code changes per week are now written entirely by Uber's internal background coding agent, and 95% of our engineers now use AI every month across all the tools we track.

This is a real reset moment for engineering; it's one of the most exciting times to lead. This shift requires builders to be curious and hands-on. I’m incredibly lucky to be surrounded by a team that’s doing exactly that.

The best part is that the strongest adoption isn’t being pushed top down from leadership announcements; it’s coming from engineers who are quietly experimenting, quietly shipping, and quietly pushing things forward.

I love spending time with those engineers because there’s no substitute for being close to the work.

Over the last few months, we leaned in hard, and the results have been phenomenal.

The bigger shift: going agentic.

84% of AI users are now working with agent-style workflows, not just tab completion. Claude Code usage nearly doubled in 2 months (32% → 63%), while IDE-based tools have largely plateaued.

Engineers are moving from accepting suggestions to delegating tasks. Even within traditional IDEs, ~70% of committed code is now AI-generated.

Background agents are writing code autonomously.

Our internal background coding agent went from <1% of all code changes to 8% in just a few months. There is zero human authoring. Engineers review and approve, but the code is written entirely by AI agents.

The role of the engineer is shifting - from writing every line to architecting systems and reviewing AI-generated code.

More to come from the @UberEng team in the coming days.

[Sergey Burykin @xtreding]

“Because if you 10x a single developer, then 10 developers can do 100x” (c)

I do agree.

[Marina Mogilko @siliconvalleymm]

🚨 “90% of code will be written by AI.”
So… is this the end of developers?

At VivaTech in Paris, I asked @github's CEO Thomas Dohmke.

His answer?
❌ “No — it’s the BEST time ever to be a developer.”

Here’s why 👇

[Sergey Burykin @xtreding]

@akhtyamovrr I’m using Cursor yes, and it’s decent 👍🏻

[Sergey Burykin @xtreding]

Everyone should use AI for coding. I’ve just build a Next JS UI MVP in a couple of hours!

[Sergey Burykin @xtreding]

It feels like everyone now builds an MCP. I hope soon MCPs will be like API - exposed by the service owner.

[Sergey Burykin @xtreding]

@george__levin Would love to hear you opinion on living in Bay Area and NYC!

[Sergey Burykin @xtreding]

@maybe_riya Tell me about famous places for sauvignon blanc Chile with their characteristics and best food pairing to it

[Sergey Burykin @xtreding]

@elonmuskADO Twitter.

[Sergey Burykin @xtreding]

@jaltma Looks awesome! I'm learning how to do latte art at home and it is a more complex task than writing code.

[Sergey Burykin @xtreding]

@uartnet Good luck!

[Sergey Burykin @xtreding]

I feel that most of people want to help others but don’t have time to answer all questions. I would like to ask successful entrepreneurs some stupid questions, and for sure can pay for it.

[Sergey Burykin @xtreding]

Roast my product idea: AI model that has been trained on your messages/emails/etc, that has same opinions as you. And public chat/API to ask questions.

[Sergey Burykin @xtreding]

@NickAnisimov 😂

[Sergey Burykin @xtreding]

V60 or espresso based coffee? 🤔
What is your preference?

[Sergey Burykin @xtreding]

@carlospaiva87 Thank you! Beautiful place!

[Sergey Burykin @xtreding]

Amazing subset in Madeira 🇵🇹

[Sergey Burykin @xtreding]

@tsilivisgeo Generally it’s not a competition! Hope you had a wonderful weekend! 😊

[Sergey Burykin @xtreding]

how is your weekend going?