// about agentpm

We built AgentPM because agentic development got fast before it got understandable.

// origin story

From agent sprawl to an evidence layer.

We started building AgentPM after spending the past year deep in agentic coding inside our AI lab. At one point, every developer on the team was managing anywhere from 5 to 12 coding agents simultaneously. At first, we were excited by the efficiency gains. Software velocity was exploding.

Eventually we started asking a simple question: what is actually happening inside these workflows? When we explored the market, most agent observability tools were focused on customer-facing AI systems like support or HR agents. Almost nothing was focused on the process of agentic software development itself.

We originally set out to build an evidence layer: a way to inspect the chain between humans and agents, including prompts, tool calls, GitHub activity, tests, PRs, issues, decisions, and outcomes. What we found was more interesting. By understanding the full path from intent to conversation to action to outcome, we could identify better workflows, surface gaps, create organizational memory, and understand how teams were actually working with agents across tools and frameworks.

Over time, AgentPM became a lightweight PM and intelligence layer for agentic development. It gives teams visibility, coordination, and confidence while creating a persistent knowledge layer of skills, tools, memory, and practices that compounds over time.

// founders

Built by operators who spent decades making complex systems observable.

AgentPM comes from the same operating belief that shaped our earlier work: complex infrastructure becomes useful when people can inspect it, control it, and improve it together.

Stephen Sorkin

Founder

Stephen Sorkin is a technology executive, inventor, and entrepreneur focused on making complex software systems understandable to humans. As co-founder of AgentPM, he is building the evidence layer for AI-assisted software development, helping teams understand what coding agents are doing, why decisions are being made, and how software evolves over time.

Previously, Stephen served as Chief Strategy Officer and SVP of Engineering at Splunk, where he helped build and scale the company's enterprise platform from its earliest years through IPO and beyond. His work spanned distributed systems, security, data analytics, and observability, with a particular focus on turning opaque systems into ones that can be understood, trusted, and controlled.

Stephen holds 148 granted U.S. patents and conducted doctoral research at UC Berkeley under renowned computer scientist John Canny. His current work centers on agentic systems, software intelligence, and human-AI collaboration.

Will Hayes

Founder

Will Hayes is a technology operator and entrepreneur with more than two decades of experience building and scaling enterprise software companies. He joined Splunk during its earliest days and spent nearly 10 years helping build the company from its first alpha release through IPO. During that time, he built and led the technology ecosystem and partnerships organization, helping establish Splunk as a leading data platform and category-defining enterprise software company.

He later served as CEO of Lucidworks for nine years, leading the company's transformation from an open-source services business into a commercial AI software platform serving some of the world's largest retailers and digital commerce brands.

Today, as co-founder of AgentPM, Will is focused on helping organizations understand, adopt, and scale AI-assisted software development. His experience building platforms, ecosystems, and enterprise software businesses informs AgentPM's mission to become the operating layer for the age of coding agents.

// what we believe

The next software stack needs a memory layer.

Evidence before opinion

Agent work should be inspectable from the original prompt through the resulting code, tests, issues, and decisions.

Memory that compounds

The best workflows, skills, tools, and project context should become durable team knowledge, not disappear into local chat logs.

Humans stay in control

AgentPM is built to help people understand, coordinate, review, and improve agentic development as the work gets faster.

// beta

AgentPM is live today.

The beta is free for teams and builders who want visibility into agentic software development.