We've spent several months building Agents, longer than any other feature at Plane. We're launching it this week. Over those months, @viharkurama and I spent hours with the team debating trade-offs and working from first principles. The hard part was fitting an agent into Plane's foundation and the way a real workspace operates. We set five requirements: 1/ Follow Plane's permission model, with access limited by workspace roles and the projects you choose. 2/ Draw from a shared workspace credit balance instead of personal AI credits. 3/ Work across supported open-weight and proprietary models. Self-hosted teams can bring their own API keys (BYOK) for supported providers. 4/ Make cloud API costs work at team scale. During the two-month launch promotion, Plane will cover most Agent usage on Cloud. 5/ Take on clearly defined responsibilities inside the team's existing workflow. I'm proud of the care the team put into v1. More phases are planned through the end of the year. Your feedback will shape the next one.

Routing isn’t the same as a Gateway. A Gateway handles auth, rate limits, observability, and request forwarding. Model routing chooses the best model for each prompt based on quality, latency, cost, and reliability. Good routing needs orchestrators, evals, and feedback loops.

Discrepancy theory studies how to allocate resources (objects, knowledge, people, vectors, or loads) as evenly as possible between two groups while balancing many attributes at once. I think someone should do this for agents, either based on single model or on multiple AI models

Individual AI = personal leverage. Team AI = shared context + lower coordination tax. The version of project management that breaks down is the one where humans are the only ones updating status and carrying context in their heads. A better system makes the work graph queryable and writable by both humans and agents, with permissions and audit logs. Issue tracking isn’t dead. Manual-only issue tracking is.

50+ Fortune 500 companies run Plane in production. Plane is the most malleable, extensible, and sovereign work platform you can deploy across your enterprise — for humans and agents. Control your data. Control your intelligence. Control your future. @planepowers for the win.

We’re hiring superstar product builders highly fluent in Al to work directly with CEO @Vamsi_Kurama here in the Bay Area. @PlanePowers has been scaling faster than any open core startup in history and not from reselling Al tokens but by deploying our enterprise product at the largest and most sensitive corporations on the planet. If you want to change the world quickly and accelerate the future of work automation, please apply or contact us — many generalist product builder roles are also open. 📈🤝

Introducing work item type hierarchy in @planepowers Enterprise Grid. Organizations can now define parent-child relationships once at the workspace level and apply them across every project. This gives teams a consistent structure to work in and agents a clearer system to reason over.

Fully agree. The AI bubble exists only in the over-hype’d and rapidly depreciating closed weights and soon to be grossly inferior compute architectures to future continuous learning compute and algorithms that more closely approximate biology.

We are not in a bubble. We are just scratching the surface. Most of the business value from AI has not been created yet. Today’s models are only the beginning. The biggest impact will come from what we build on top of them and from the new ways they change how we work, learn, discover, and create. There is still an enormous amount of work to do. All we need is openness and maximum acceleration. Progress has always come from people sharing ideas, building on each other’s work, and moving faster together. We should focus on the fundamental units of progress. • Compute maxxing. Making computation cheaper, faster, and more accessible. • Token maxxing. Giving intelligence more room to reason, understand, and communicate. • Tool maxxing. Building better ways for AI to interact with software, data, and the real world. • Knowledge maxxing. Making the world’s knowledge easier to access, organize, and build upon. • Human maxxing. Helping every person learn faster, build more, and solve bigger problems. Every improvement compounds. Better infrastructure leads to better models. Better models lead to better products. Better products lead to better organizations. Better organizations lead to faster scientific discovery and a better quality of life. The goal is not to maximize valuations or benchmarks. Those are outcomes. The goal is to maximize human progress. If we keep improving the fundamental units of progress, everything else follows. We are only scratching the surface.

We are now enabling work item type creation at the workspace level in @planepowers on our Enteprise Grid plan. This may sound like an administrative change, but it addresses a problem that becomes increasingly important as an organization adds more projects. Work item types define the information teams capture for different kinds of work. A Bug might require severity and environment. An Incident might require impact and root cause. These definitions shape how work is created, filtered, automated, and reported. More on 🧵