Agentic coding adoption at Airbnb

AI generated summary of the linked source. Posted anonymously. Transcripts are published only for public links, never for uploads, documents or pasted text.

Your library

The AI link is plain markdown. Paste it into any assistant and it can read the whole public entry.

Summary

Two Airbnb developer platform engineers, Szczepan and Mike, describe how the company bet on agentic coding in early 2025, using a pair programming analogy to explain the shift to leadership. They walk through adoption data, a four level skill model for engineers learning to trust and steer AI agents, and the infrastructure (sandboxed workspaces, parallel sessions, code review culture) needed to make it work.

Core ideas

One context card plus 9 ideas

Bold vision over incremental steps: Airbnb set an ambitious 2025 target for agentic coding rather than planning small improvements, because big step changes need a destination to work backward from.
Pair programming reframed: the developer becomes a permanent navigator steering an agentic session while the tool acts as the tactical driver materializing code changes.
Predictions beat expectations: Airbnb forecast 20 to 40 percent of PRs using agentic sessions by end of 2025; they already sit in the sixty-plus percent range.
Success measured holistically: predictions are for leadership buy-in, not KRs. Real success tracking combines developer sentiment surveys, tool adoption curves, and objective engineering metrics like PR velocity.
Agentic loop defined: an engineer specs or prompts, then a loop runs that can call LLMs multiple times, invoke MCPs and tools, load config (system prompts, agents.md, cursor rules), with output requiring human review before a PR goes out.
Four-level skill journey: from approving every AI operation, to auto-approve once prompting skill improves, to running parallel agentic sessions, to building and integrating the tools themselves.
Trust the skill, not the AI: Szczepan says he trusts his own prompting ability, not the model itself, and stresses pushing back on suspicious agent output until it admits mistakes.
Parallelization needs infrastructure: sandboxed cloud workspaces (Airbnb built "AirDev Workspace") beat local tricks like Git worktrees for running multiple agentic sessions reliably.
Review culture becomes the job: as agents write more code, test code quality and code review discipline matter as much as production code, possibly becoming the main engineering skill.

Quotes

“I can't believe that any of this made it easier, but AI allows people like me, I'm a college dropout.”

“You ask it, right? Like, "Why did you do this in line 120?”

“Maybe software engineering of the future is, like, all a, it's, it's all code review, right? I'm okay with that, so.”

“We don't force engineers to use it. It's a supplemental tool.”

“Getting too fast to putting your tool on auto-approve is not great.”

Resources

Links to the tools, products, repos or reading this entry mentions. Found by a web search of what the entry names, so you can go and use them.

Anyone can run this once. After that the links are saved on this entry and shown to everyone.

Remix this knowledge

Turn this entry into something you can post. Pick a format and get a fresh, original rewrite.

Remixes are AI generated, original wording, and free to reuse.

Talk it through

Open a round table on this entry and argue with other people about what it means.

Start a round table

Drop your own

Paste any video, podcast or audio link and get the transcript, the core ideas and the quotes back in one pass.

Try Drop