We just shipped a one-line change that doubles the default budget for our sc-qa agent from $1.00 to $2.00 per task. On the surface, it's a tiny tweak. But it reflects something important about how we think about quality and how autonomous agents scale.
The Problem
SC-QA is our quality assurance agent. It runs test suites, validates deployments, and catches regressions before they reach production. Over the past few weeks, we noticed a pattern: complex test scenarios—especially those involving multi-step integration tests or extensive E2E validation—were hitting budget limits mid-execution.
When an agent hits its budget ceiling, it stops. No partial credit, no 'close enough.' The task fails, and we lose visibility into what would have been caught. That's not acceptable for a QA workflow.
The $1.00 default made sense when sc-qa was primarily running unit tests and quick smoke checks. But as the platform matured and test coverage deepened, the budget became a constraint rather than a guardrail.
The Fix
We bumped the default to $2.00. Simple as that. The change was made in anti_strug/agent_registry/config.py, the central configuration file that defines runtime parameters for all Strug Works agents.
This gives sc-qa enough runway to complete thorough test passes without constantly bumping into artificial ceilings. It's still a budget—agents don't get infinite resources—but it's calibrated to the actual work they need to do.
Why This Matters
Agent budgets aren't arbitrary. They're operational boundaries that shape how autonomous systems behave under resource constraints. Set them too low, and you create failure modes that have nothing to do with code quality. Set them too high, and you lose cost predictability.
We optimize for completeness in QA workflows. A test suite that runs 80% of the way and fails due to budget exhaustion is worse than no test suite at all—it creates false confidence.
This change is also a signal about platform maturity. As our test coverage expands and our validation workflows grow more sophisticated, agent budgets need to scale with them. We're not building throwaway scripts; we're building durable infrastructure.
What's Next
We're monitoring sc-qa's actual budget utilization patterns now that the ceiling is raised. If we see consistent headroom, we'll know $2.00 is the right default. If we see new budget failures emerge as test complexity increases, we'll revisit again.
Longer term, we're exploring dynamic budget allocation—agents that can request additional resources mid-task if they hit justified complexity thresholds. That's a harder problem (it requires trust models and abuse prevention), but it's the right direction for truly autonomous systems.
For now, sc-qa has the budget it needs to do its job well. And that's exactly how it should be.