Kein Burggraben, nur Execution: enact Pre-Seed, Physical AI & Lette – Constantin Sanders (Peak)

Peak backs AI work replacement, where only execution speed counts

Peak is betting that AI products will replace human work over time, in software like Lette and in hardware like enact. With no lasting moats, early backing comes down to the founder and how fast the team executes. Sanders expects the next one to two years to show how reliably this works, and enact is still far from replacing a whole person.

Takeaways

• Lette automates property management at real scale. Lette runs on more than 30000 apartments and already automates over 60 percent of incoming requests, from rental applications to broken boilers. It integrates with systems such as Yardi, and JLL is a customer. Workflows leave little room for hallucinations.
• Moats are gone, so execution and speed decide. Sanders says a durable moat hardly exists today. US player Elise AI has raised more than 100000000 for what Lette builds in Europe, yet Peak doubts it will enter soon. Peak watches how fast Lette wins large customers and expands within them.
• enact is a bet on founder and topic. The Zurich company raised a 30000000 dollar pre-seed from Red Alpine and Notion Capital, with angels including Yann LeCun. Founder Eve Gallika led software on 1X's NEO humanoid and worked at Physical Intelligence. There is no paying customer yet.
• Grip and touch still block factory robots. Robots see well but cannot judge grip force, so assembly tasks such as fitting door foam at car makers like Mercedes still need people. enact works on hands and motors that sense force. By one statistic, robotics solutions are 50 to 60 times the software market.
• Robotics money is huge but results lag. About 18800000000 dollars flowed into robotics in 2026 up to early July, yet the ChatGPT moment is still awaited. Sanders says the deployment layer decides outcomes, and results at scale have disappointed on both speed and accuracy. Peak avoids classic humanoid makers because hardware is so capital intensive.
• Revenue metrics need careful scrutiny. Annualized run rate is the last quarter times four or the last month times twelve, which differs from contracted ARR. Sanders says micro1's figures were sometimes equated with ARR defined differently. Net revenue retention was floated as an alternative, and Peak digs deep in due diligence.

Peak's four-layer robotics map

• AI models as robot brains, including Physical Intelligence, Skild AI, Google DeepMind, NVIDIA and Figure.
• Training data, where Scale AI and 1X are the two big players and micro1 raised a 55000000 seed round.
• Hands and touch sensing, led in Europe by enact and Mimic Robotics, both in Zurich, while Tesla and Figure work on it in the US.