Decoding AI’s Open-Source Course Maps Three Ways to Run an Agent Loop and the Provider Economics Behind Each
Most teams treat ‘which model’ as the important decision. The harness engineering literature keeps pointing somewhere else. In LangChain’s Terminal-Bench experiment, changing only the harness—same model throughout—moved a coding agent from roughly 30th place into the top 5. That result reframes the question. If the harness decides quality, then how you run the loop becomes an architecture decision, not a deployment detail. Paul Iusztin’s open-source course Building a
Make your prediction
Will a coding agent harness paper from LangChain be cited in a new study before September 10, 2026?
Resolves by Sep 10, 2026
Your prediction
50% · 50/50 coin flip
NOYES
Get smart on it
An open-source course demonstrates three different ways to run an agent loop, with each approach suited to different operational needs and cost structures. The key insight is that how you architect the system matters more than which model you choose, as swapping only the harness moved an agent from roughly 30th place to the top 5 in testing. Interactive mode, where a human watches in real time, works best with per-token pricing from hosted APIs because latency matters most. Remote and async modes, where the system processes work without constant human observation, work best with per-GPU-hour billing because throughput and efficiency matter most, making batch processing on remote servers significantly cheaper than API calls at scale.
Related news
Runable hits $21M to bet AI agents can go from building businesses to growing them
Runable says 60%–70% of its 1 trillion-plus token usage in the last 90 days came from paying customers.
How loveholidays is making everyone a builder with Codex
Discover how loveholidays uses OpenAI Codex to make software development accessible across the business, helping teams turn ideas into products faster.
