TL;DR: Sona8 maps company’s workflows with the employees who know them best. Employees explain and show their day to day work to Sona8’s agent: in a conversation, on a screen share, through the phone camera, or by letting Sona8 record the steps in the background while they work. This produces a concrete records of processes used to fix inefficiencies and guide AI agents to automate them.
🧩 The problem
The bottleneck for automation is domain knowledge. LLMs have the reasoning capabilities, but they lack the operational context. E.g., how a quote gets approved, which exceptions matter, who gets called when the system is wrong; this knowledge lives in people’s heads, in email threads or old excel files. People remember it, but agents can’t read it. So automating one step creates new problems next to it, because the automation doesn’t match how the work is done.
🛠️ What we’re building
Employees explain and show their day-to-day work to our agent, in a conversation, on a screen share or through the phone camera. It asks follow-up questions where something is unclear and produces a map of the workflow as it runs today, exceptions included. Teams pull these maps into Claude, ChatGPT or their own agents via MCP to improve processes and give AI agents instructions.
🆕 What’s new
Employees can now share their screen, point a phone camera at physical tasks, or let our agent record the steps in the background while they work.
📈 Where this goes
👋 Team
At BCG & McKinsey, we helped companies map their processes and identify what to improve. The answers were always with the people doing the work, rarely in the documentation.
If you’re running a transformation or AI rollout (at your own company or for a client), talk to us.
Thilo and I worked together at BCG, interviewing employees and repeatedly going back to them during implementation. On our evening runs, we started asking whether AI could make those conversations possible across an entire company.
We started building after BCG, while I was at MIT and Thilo was researching conversational AI at TUM. As pilots gained traction, we both left our programs. Madeleine brought her experience as a self-taught engineer, repeat founder and former principal engineer at McKinsey QuantumBlack.
Every major decision depends on context. Much of it sits with employees who know how work happens and why things fail, but that knowledge rarely reaches the people deciding what to change.
We’re building Sona8 into the operating system for strategy and transformation teams. We will combine what our voice agents learn from employees with company data to build a live model of the organization. Teams will use it to compare alternatives and understand what a decision would require across sites and departments.
The same model will be the context layer for AI automation. Sona8 will use it to deploy agents and coordinate execution, then learn from the results. Small teams will be able to steer company-wide change, with the organization keeping what it learns as people and agents move on.