Ona

Run a team of AI software engineers in the cloud.

[OpenAI Newsroom @OpenAINewsroom]

We’ve reached an agreement to acquire @ona_hq.

Its secure cloud execution technology will help Codex take on longer-running work, even when laptops are closed, and help more organizations deploy agents securely in production.

After closing, Ona will join OpenAI’s Codex team. https://t.co/RXoL4LSDcf

[Ona @ona_hq]

https://t.co/dJAkZScksZ

[Ona @ona_hq]

375 PRs merged in 10 days. 87% of them merged autonomously, with no human taking any action.

Zach Malguitou spent 10 days building a "software factory" in public, livestreaming the setup of 16 background automations building a real product.

The output was 68,000 lines of code and over 1,000 tests. Issues were closed on average in 38 minutes, PRs merged in under 5 minutes, and 98% of agent executions triggered by the factory itself with no human in the loop.

Full session recording link is in the comments!

[Ona @ona_hq]

https://t.co/dJAkZScksZ

[Ona @ona_hq]

There are three actors on every engineering team now, and most companies are still only designing for two.

For years the picture was simple: humans use your product, your product responds. Lawrence Jones from https://t.co/tWwwOyrYSG explained at the Background Agents Summit why that picture is outdated.

There are humans, the central specialist agents companies build (https://t.co/MllBzngHSa's AI SRE for production issues, Cursor for code changes, Fin for support), and the personal agents that each human is now paired with. All three need to hand off work to each other.

The MCP contract between your agents has become one of the most important UIs your company ships. It's the surface your agents see of each other, in a way most APIs never had to be.

Full session recording link is in the comments.

[Ona @ona_hq]

https://t.co/13uiLxLpWc
https://t.co/j3rJMmy69F

[Ona @ona_hq]

What is the best first use case for background agents?

Four filters for determining a good first use case:

- High-volume (savings compound)
- Well-defined (agent can verify its own work)
- Low blast radius (failure is recoverable)
- Measurable (real before-and-after numbers)

Six proven starting use cases with ready-to-use templates are available in the guide and templates library in the comments.

[Ona @ona_hq]

https://t.co/dJAkZScksZ

[Ona @ona_hq]

Most regulated companies hear "AI" and reach for a blanket ban. Suhail Patel from Monzo (a regulated UK bank) said it well at the Background Agents Summit:
"There's nothing in regulation that stops you from adopting AI."

Regulation gives you guardrails and penalties for crossing them. It was never designed to block new technology.

The actual risk is specific. Connect your data warehouse to a model that trains on its inputs and your customer data just got exfiltrated into someone else's training set. That's where penalties land.

Monzo's approach: enable the tools, put strict controls on what they can ingest.

Full session recording link in the comments!

[Ona @ona_hq]

Read more and determine where you stand at: https://t.co/g28Tz9Ld6M

[Ona @ona_hq]

"We're already doing AI. Our devs use Claude."

Has your team shipped more actual features and is your backlog shorter?

Three phases:

1. Code assistants (individual speed)
2. Background agents (org throughput)
3. Software factory (autonomous SDLC)

Most teams are stuck at phase one thinking they've arrived.

[Ona @ona_hq]

https://t.co/dJAkZScksZ

[Ona @ona_hq]

Every major model has written solid code for a well specified task for over a year.

So why are most companies still reviewing every diff while a handful ship autonomously? Shardul Vaidya from AWS at the Background Agents Summit: it comes down to verification.

Link to the full session recording in the comments!

[Ona @ona_hq]

Six teams built background agent infrastructure independently. All converged on the same five primitives:

- Sandboxed dev environments
- Context and connectivity
- Triggers
- Fleet orchestration
- Governance and security

These are a stack and the order in which you put them into place matters.
Learn why at: https://t.co/ChU3eTZxuY

[Ona @ona_hq]

https://t.co/dJAkZScksZ

[Ona @ona_hq]

Engineers are becoming production line managers.

Cole Murray at the Background Agents Summit: we're moving from software engineering to software manufacturing. The artisan model (one engineer, one task, handcrafted code) is giving way to fleets of background agents shipping the work.

Link to the full session recording in the comments!

[Ona @ona_hq]

https://t.co/7vW8oM0X8m

[Ona @ona_hq]

Stripe and Ramp built their own agent infra. But they had cloud dev environments for years before agents were even a thing.

If you're starting from scratch, here's the actual shopping list. Six systems, each with ongoing maintenance:

- Sandboxed dev environments
- Agent orchestration
- Trigger system
- Agent harness
- Observability
- Security and isolation

We dive into the full details in section 4 of the "Engineering Leader's Guide to Background Agents," link in the comments! ⬇️

[Ona @ona_hq]

https://t.co/dJAkZScksZ

[Ona @ona_hq]

At Stripe, designers and PMs are merging PRs without pulling in engineers.

Alistair Gray broke this down at the Background Agents Summit: a PM spots a small thing that could be better, hands it to a Minion agent, gets a branch and a PR back, and only loops in an engineer for the final review.

Why Stripe built it: employee attention is the most precious commodity they have.

Link to the full session recording in the comments!