Matt Turck interviews Basis co-founder on long-horizon agents: behavior specs, ontologies, process supervision
schwentker · x · 2026-09-03
Matt Turck talks with Mitch Troy, co-founder of Basis, about building reliable, trustworthy long-horizon AI agents. Basis focuses on accounting and runs autonomous multi-day tax returns.
Key topics:
- What makes an agent truly long-horizon: behavior specs, ontologies, and process supervision
- Accounting framed as "an intelligence over the economy" as an ideal agent domain
- Agents that hand off work like senior engineers
- A brief history of agents from ReAct to today; why AutoGPT fell short; why LLMs lack long-term memory
- Three breakthroughs — Opus 3, o1, o3 — and how reasoning models unlocked agentic capabilities
The core thesis: agent reliability comes not from smarter models but from architecture built around specs and process supervision.
More from coding & agent
- Devs say LLMs are over-optimized for one-shot answers and refuse to ask for feedback mid-task — Elijah_Meeks · 2026-09-03
- Namespace partners with Cursor to give cloud agents native Mac/Linux Devboxes — dean_rie · 2026-09-03
- Open-Source MCP Tool Bridges iOS Simulator Context to Coding Agents — ivanzhaowy · 2026-09-03
- Meta's CORAL: An LLM-Native Harness That Continuously Optimizes Production Recommenders — _reachsumit · 2026-09-03
- My Allowlist Was Cited in Five Design Docs and Never Read at Runtime — Thirumalaiboobathi · 2026-09-03
- Developers are replacing --help with atuin's AI for unfamiliar shell commands — braelyn_ai · 2026-09-03