@RobGibson20 In June, we released AutoPM as part of a research preview. Our first multi-agent innovation intelligently decomposed complex tasks into structured coding subtasks, planning and executing each component independently.
We brought you an autonomous product manager for your most ambitious development projects.
Today, we’re taking a much bigger step forward.
Cosine is now the first AI software engineer in the world powered by GPT-5.
With this leap, we can fully integrate our multi-agent capability into Cosine’s core reasoning, eliminating the need to toggle it on or off.
From research to integration
AutoPM brought multi-agent reasoning to our platform, enabling Cosine to assign, coordinate, and manage tasks like a real product manager would. Early adopters saw how powerful that could be. AutoPM quietly optimised dozens of coding sessions, jumping in when needed, stepping back when not. From Slack triages to Linear tickets, teams used AutoPM to move faster and stay in flow.
We listened closely to the feedback.
Users told us AutoPM worked best when it felt invisible; when it made decisions, delegated work, and orchestrated multiple agents without fanfare. Our power users leaned in hard, using AutoPM across complex workflows where a single-agent approach fell short.
Now, multi-agents are fully native to Cosine’s experience, and that experience just got a massive upgrade.
Every coding task is now powered by GPT-5. That means stronger reasoning, tighter coordination, and significantly faster execution across the board.
The world’s first engineer powered by GPT-5
Cosine’s transition to GPT-5 marks a new era in AI-assisted development. Improved reasoning models mean you’re working with a teammate capable of expert-level decisions, context-aware planning, and multi-agent execution.
GPT-5 dramatically boosts Cosine’s reasoning abilities, memory coherence, and task decomposition performance. With the seamless integration of multi-agent reasoning, Cosine is now more autonomous, more capable, and more enterprise-ready than ever before.
Smarter coordination, zero config
Modern software development is inherently collaborative. Tasks aren’t linear, and developers often juggle multiple concerns: testing, implementation, documentation, integration.
Multi-agent reasoning lets Cosine mirror that complexity with intelligent delegation. You get the benefits of a product manager without adding another teammate.
And because multi-agent reasoning is now built-in, you don’t have to think about it. Cosine just does the right thing, whether that’s spinning up a helper to write tests or coordinating a thread fix alongside a feature push.
With GPT-5 at its core, Cosine makes those decisions faster and with deeper accuracy. You’ll see it in how it asks better questions, prioritises better actions, and adapts more fluidly to your workflow.
"Quick" vs "Think" mode
Alongside this release, users will notice some new options for task delivery.
"Quick" is your lean, turbo-charged reasoning path, and "Think" is a comprehensive, multi-step reasoning engine.
Either way, multi-agents are ready and waiting to dynamically deploy behind the scenes, delivering outcome-driven results.
A new standard
Enhanced by GPT-5, our general release of multi-agents simplifies the user experience while making Cosine more powerful under the hood. We’ll continue tuning our multi-agent capability to be even more adaptive, learning from how teams work and optimising coordination patterns automatically.
For our enterprise users, this change reflects a broader principle: intelligence should feel effortless. Your coding agent should manage complexity, not expose it.
No switch to flip, no toggle to find. Cosine just became more powerful and coordinated than ever before.