cd /news/ai-agents/your-agent-keeps-re-deciding-settled… · home › topics › ai-agents › article
[ARTICLE · art-168384] src=ainexusdaily.vercel.app ↗ pub= topic=ai-agents verified=true sentiment=· neutral

Your agent keeps re-deciding settled questions. Give each decision a scope.

The AuzzurA team published a Dev.to post arguing that coding agents re-litigate settled questions because decisions live in unstructured rules files rather than scoped entries, and proposing a five-field decision format: Statement, Scope, Status, Replaces and Why. Under the approach, only Accepted decisions whose scope matches the task's files or task type go to the top of the prompt, and agents write new findings as Proposed entries for one-time human review. The post promotes AuzzurA's DM product, which stores approved Rules, Skills and Decisions and returns agent learnings for review.

read2 min views1 publishedNov 12, 2026
Your agent keeps re-deciding settled questions. Give each decision a scope.
Image: Ainexusdaily (auto-discovered)

If your coding agent keeps reopening questions you already settled ("should we use the ORM here?", "can I change this public API?"), the problem usually isn't the model. The decision exists only in someone's head, an old chat, or a long rules file where it competes with fifty other lines. A rules fi le grows by addition. Old guidance stays next to new guidance, nothing says which part of the codebase a line applies to, and the agent gets all of it on every task. It follows some lines and quietly drops others. A decision your agent can use has five fields: Statement.

One or two sentences, phrased as what to do or not do. Scope. Which directories, services or task types it applies to. Without a scope it gets applied everywhere, or nowhere.

Status. Proposed, Accepted or Superseded. Only Accepted decisions go to the agent. Replaces.

The earlier decision it supersedes, so the old one stops reaching new tasks. Why. One sentence. It helps the agent handle cases the statement doesn't cover.

Before a task, select the Accepted decisions whose scope matches the files or task type, and put those at the top of the prompt. Ten relevant decisions beat two hundred lines of history. At the end of a task, ask the agent: "Did you make or discover any decision the next task should know? Write it as a Proposed entry." A person reviews it once.

After that it's reused, without asking again on every task. With several people and agents, the file becomes the bottleneck. That's the problem we work on at AuzzurA with DM: it keeps the approved Rules, Skills and Decisions, gives each task the ones that apply with the reason each applies, and brings the agent's learnings back for review. https://dm.auzzura.com?utm_source=devto&utm_medium=social&utm_campaign=dm_launch Disclosure: this post is from the AuzzurA team, which builds DM.

Key Takeaways #

  • •Old guidance stays next to new guidance, nothing says which part of the codebase a line applies to, and the agent gets all of it on every task.
  • •This story was reported by Dev.to , covering developments in thedev space.
  • •AI advancements continue to reshape industries — read the full article on Dev.to for complete coverage.

📖 Continue reading the full article:

Read Full Article on Dev.to →

── more in #ai-agents 4 stories · sorted by recency
── more on @auzzura 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/your-agent-keeps-re-…] indexed:0 read:2min 2026-11-12 · —