Announcing the background agents landscape

TL;DR: We published an interactive landscape of background agent infrastructure, with 103 tools across 11 layers. It's a map for teams inspired by the stories of Ramp, Stripe and others, to help build their own background agent infrastructure. We're also now planning a virtual summit to bring together like-minded companies to shape the future of the SDLC: sign up to get notified.

Introducing the landscape

Engineering teams are watching Stripe merge a thousand agent-authored PRs a week, Ramp attribute over half their PRs to agents, and Spotify ship 1,500+ agent-generated changes into production. The natural question is: how do I do that?

When we launched background-agents.com, these big tech use cases were all published at roughly the same time. The response to the microsite told us we'd hit a nerve. Many people told us that the site had put into words (or rather visualizations!) of what they were feeling and experiencing with coding agents.

But the follow-up questions we received followed a similar pattern: "I am personally sold on the idea of background agents. My CTO also wants us to get to 100% agent driven commits ASAP, where do I start?". Today, we're helping you get one step closer to that answer with the background agent tool landscape.

Who is the landscape for?

Stripe, Ramp, and Spotify all built their agent infrastructure on top of years of existing investments. Stripe had cloud devboxes before GPT-3 existed. Spotify's Fleet Management system was running before they added agents. However, most companies don't have that luxury. The landscape is for the teams that need to assemble a background agent stack from the ecosystem, which is most of us.

What we learned building it

In creating the landscape, we naturally had some observations. Here are just a few, I'm sure we'll share many more in the coming weeks as the landscape evolves.

The agent layer is well understood, orchestration is not. There are currently 16 tools in the agent layer, and we could have listed 100 more. However the industry is also converging towards a handful of base agents/harnesses like Claude Code and Cursor. The real challenge now, is not with the harnesses themselves, but with the orchestration of large-scale fleets of agents. Agent offerings are now expanding to include workflow and orchestration like Cursor and Ona Automations.

Sandboxes and dev environments are not the same thing. There's also a lot of category confusion, most notably with sandboxes. Sandboxes are a category that started out primarily to add code execution to agentic software like ChatGPT or Claude. But, is now being re-purposed as more general purpose development environments for agents, despite not being fit for purpose. We expect more granular terms to emerge to meet the growing need for different sandbox types.

Security and identity is the biggest gap. Security challenges are one of the main areas holding back mass-market adoption of background agents. Tools like Keycard are emerging to tackle challenges of agent identity. Security functionality is also tightly coupled to the runtime layer. Ona's Veto, for example, enforces policies at the kernel level to address the issue that path-based denylists don't survive an agent that can reason. But every platform approaches security differently. Industry collaborations like REBAR are likely to help here.

Real co-ordination for the agentic SDLC doesn't fully exist yet. The choice today for background agent co-ordination is mostly to extend existing tools like GitHub using webhooks and git events. Projects like OpenAI's Symphony give us a glimpse into the future. Yet, many of these projects are not yet production-ready. Only yesterday, we saw early signs of issue-trackers like Linear now taking steps towards becoming a co-ordination layer, and other developer tools like JetBrains experimenting in a similar direction.

Virtual Summit, coming soon

In addition to the landscape, we're now also preparing for a virtual summit - planned for the end of April. To get organizations who are building background agent infrastructure into the same room, share what's working and figure out the future of background agents and the AI SDLC. If you want to join, speak, or run a workshop, you can find the register interest form on background-agents.com.

Tell us what you think

We expect the landscape to look very different in six months. If we're missing something, or if you disagree with how we've drawn the lines, we'd love to know. Check it out, share it with your friends and colleagues, we hope it helps!

background-agents.com/landscape

Is this landscape vendor-neutral? That's the goal. If you spot a gap, let us know.

How often is the landscape updated? Regularly, as the ecosystem evolves.

How do I suggest a tool? Message Lou on X or LinkedIn.

Is the landscape open source? Not yet, but we're working on it.

Can I use this internally? Yes, and we made a slide deck you can use.