AI decision governance
Runtime guardrails decide what an agent may do. Governance decides whether what it did was right, and whether the team learned from the last time a person disagreed.
What AI decision governance is
AI decision governance is the practice of making the business decisions behind an AI workflow explicit, reviewable, and consistent. It sits above guardrails. A guardrail decides what an agent may do. Governance decides whether what it did was right.
Solon is that governance loop. It records every decision and every override, finds the repeated overrides that signal a missing rule, and turns them into policy your team can review and merge.
Why now
The EU AI Act's transparency obligations have applied since August 2, 2026, with broader high-risk obligations phasing in around the same window. Regulators, auditors, and enterprise buyers are converging on one expectation for AI-driven decisions: decision-level evidence, with human oversight that is visible and traceable. Activity logs show that an agent ran. Decision governance shows what was decided, who overrode it, and what changed as a result, which is the evidence that survives an audit.
How it differs from the tools you already have
- Agent runtimes and access control decide what an agent can reach. Solon decides whether the outcome was right.
- Evaluation and observability suites score outputs after the fact. Solon learns from the overrides a person already made.
- GRC platforms document the organization for auditors. Solon produces executable, versioned rules.
- Static policy engines such as OPA and Rego run rules engineers author by hand. Solon drafts rules from real decisions.
Why overrides are the raw material
Every override is a manager paying attention. Someone looked at a case, disagreed with the system, and made a different call. That disagreement is the honest signal a business has about where its policy is wrong or missing.
One override is noise. A cluster of similar overrides is a missing rule. Three similar cases is enough for Solon to propose one.
The loop, in brief
Solon wraps one decision function in your app. Decisions and overrides are recorded append-only, with PII redacted per your policy. Repeated overrides cluster in a rolling window. Solon drafts a rule and opens a pull request against your policy repo. A human reviews and merges. Every future decision resolves with the merged rule.
The thinking behind Solon is public: principles, architecture, decisions, and the build log live at github.com/rubenabarron/solon-open.
Keep reading
- How Solon turns overrides into policy: Capture every decision, cluster repeated overrides, draft a policy pull request, and keep the merge a human decision. The loop, step by step.
- Refund and exception decisions, made consistent: Goodwill refunds and exceptions vary by manager and shift. Solon makes the precedent explicit, versioned, and consistent across every reviewer.
- Frequently asked questions: What AI decision governance is, how Solon learns from human overrides, what a design partnership looks like, and what stays a human decision.
- Work with us: A few teams running refund or exception decisions work with us directly. You keep your policy and evidence; your feedback shapes what we build.
- Built in the open: Solon's product principles, architecture, dated decisions, and build log are public. Curated in the open; every entry is merged by a human.
- Privacy note: What the Solon design partner form collects, why, where it lives, and how to have it deleted.
- See the loop on a real case: Walk one recorded refund case end to end: a System One model proposes, Solon escalates, you override as the manager, and three similar overrides become a policy pull request.