- Author
- brian
- Published
- 07 May, 2026
- Kind
- Very short personal blog post, argument only
- Original
- bsuh.bearblog.dev/agents-need-control-flow/
- Subject
- Harnesses and agent loops
- Themes
- Handing over control, Checking the work
What it is
The author signs as brian and writes a short thesis on reliable agents. The post sets out to say why prompts alone cannot carry complex tasks.
What it says
- The post argues reliable agents need deterministic control flow in code. Longer prompt chains are not enough.
- The author says MANDATORY or DO NOT SKIP show prompting has hit its ceiling.
- The post compares prompt chains to a language where statements are mere suggestions.
- It proposes scaffolds with explicit state transitions and validation checkpoints. The model is one component.
- It says an agent without error detection quickly reaches wrong conclusions.
- Without programmatic verification, the post says, the options are babysitter, auditor or prayer.
What evidence it gives
The post offers argument only. It has no code, benchmarks or worked examples. The page is very short.
Interests and limits
The post is one person's opinion and gives no test of its claim. It does not say how much control flow is enough. It does not name any product or employer with a stake in the view.
Connections
These links are this wiki's own reading, not claims the sources make.
- Finite State Machines in Forth. The blog post wants explicit state transitions, with the model as one component. Noble's old note shows what that can look like: a table you can read. Noble also makes each transition a word, which permits nondeterministic machines. A model call could be one such word.