Kam Lasater

Kam Lasater is a software engineer and startup founder who cofounded SeeClickFix and Cyclic before turning to practical AI agents and model evaluation. His work spans civic technology, serverless developer infrastructure, and software that makes autonomous systems easier to inspect and control.

Lasater studied electrical and computer engineering at Carnegie Mellon and began working at Higher One in 2003, building financial-technology products as a software engineer and architect. In 2008, he cofounded SeeClickFix, a platform through which residents report neighborhood problems and local governments manage service requests. CivicPlus acquired the company in 2019.

He subsequently led LegalScience, served as chief technology officer at HeavyWater, and became a senior development manager at Black Knight following HeavyWater’s acquisition. At HeavyWater, he worked on AI systems for classifying and extracting information from mortgage documents.

In 2021, Lasater cofounded Cyclic, a serverless hosting platform designed to spare JavaScript and Python developers the infrastructure complexity involved in deploying applications. His account of starting Cyclic describes the disproportionate operational burden placed on small applications. He closed the company in 2024 after sales proved insufficient to sustain it.

At AI Engineer Summit 2025, Lasater represented Manufactured, where he worked on operational AI systems for tasks including sales, underwriting, vendor sourcing, and bookkeeping.

• Framework-free agent architecture: His agent-building demonstration and public example code assemble an agent incrementally from model calls, structured evaluation, web search, planning, writable task state, and an execution loop.

• Application-owned tool execution: The model selects tools and proposes their arguments; application code performs the calls, returns results, maintains control flow, and handles parallel execution, schema failures, and runaway loops. A simple to-do list supplies persistent planning and task state, while a second model evaluates whether the original assignment is complete.

• Personality Benchmark: Lasater tested 15 language models with the PID-5 personality assessment, reporting that nine crossed thresholds used in human screening. His Personality Benchmark project explores behavioral tendencies in AI assistants without claiming that models can receive clinical diagnoses.

• Accessible AI-assisted development: Lasater encourages junior developers to treat coding assistants as an opportunity, comparing their emergence with computer-aided design in architecture. His YOLO agent sandbox provides a disposable containerized environment for experimenting with AI coding tools.