We just merged Wave 1 of a multi-phase architecture upgrade that touches the core of how Strug Works agents think, remember, and coordinate. This isn't a flashy feature launch—it's foundational work that makes everything else possible.
What Changed
This release covers Linear issues SCE-146, 148, 149, and 150—each targeting a different layer of our agent intelligence stack. At the highest level, we've introduced Phase 2 memory architecture that allows agents to store and retrieve context more deliberately, and Phase 3 routing logic that helps tasks find the right agent with better precision.
Previously, agent memory was reactive—data was logged, but retrieval was basic. Now, agents can scope memory by task, role, or globally, and query with confidence scoring. That means an sc-backend agent working on a new API endpoint can pull relevant patterns from past work, not just generic logs.
On the routing side, we've added intelligence layers that evaluate task context before dispatch. Instead of simple keyword matching, the system now considers agent workload, specialty alignment, and historical success rates. It's still early—Wave 1 is about getting the plumbing right—but the impact is already measurable.
Why It Matters
Autonomous agent teams live or die by context. A human engineer knows to check Slack, read the last PR, and ask a teammate before starting work. Agents need equivalent affordances, and that's what this architecture unlocks.
Better memory means fewer redundant decisions. Better routing means less thrashing and faster time-to-merge. For customers using Strug Works, this translates to more consistent output quality and reduced token costs on tasks that previously required multiple retries.
What's Next
Wave 1 establishes the primitives. Wave 2 will focus on cross-agent memory sharing and dynamic routing policies—letting agents learn from each other's work and adapt task assignment rules in real time.
We're also building observability tooling so teams using Strug Works can inspect memory scopes and routing decisions through Strug Central. Transparency into how agents think is critical as we scale autonomy.
This is infrastructure work, but it's the kind that compounds. Every agent interaction from here forward gets smarter. That's the bet we're making with Phase 2-3, and Wave 1 is the first proof point.
Commit: 8267ae6 | Author: strugcity | Issues: SCE-146, 148-150