Tracecat hit 1.0 GA with an open-source agent stack this week, while Fibery rebuilt its AI agent from scratch — both reflect a PM sector reckoning with what agentic architecture actually requires.

The week in project-management

Two kinds of architectural reckoning defined project management this week. The first: shipping 1.0 GA with honest posture about what's open and what's proprietary. Tracecat open-sourced its agent stack and hit 1.0 stable in the same release. The second: acknowledging that the first AI agent architecture didn't hold up. Fibery rebuilt its AI agent from scratch — replacing a router-over-sub-agents model with a single dynamic-tool-loading agent. Both moves signal the sector is past the "ship AI features" phase and into the "make them actually work" phase.

On the scheduling and resource side, Float added AI experience profiles — a move toward answering "who has done this before" rather than just "who is available." Project management is slowly absorbing the resource intelligence that was previously trapped in tribal knowledge.

Leaders

Tracecat 1.0.0 hit general availability with three simultaneous moves: SSRF-hardened MCP connectivity for agent-to-external-service calls, open-sourcing the agent component stack (presets, skills, and catalog connectors), and self-hosted model support. Going 1.0 stable while simultaneously open-sourcing the agent layer positions Tracecat for enterprise adoption — regulated industries can audit the stack. The company started the 1.1.0 alpha the same day.

Fibery replaced its AI agent entirely. The prior model — a router dispatching to specialized sub-agents — accumulated context loss across handoffs and was slow. The replacement is a single dynamic-tool-loading agent: one context, tools loaded on demand, no sub-agent boundary to cross. Whiteboard read/write and a model picker shipped alongside the new architecture. Fibery is the clearest example this week of a team treating their AI agent as a first-class engineering problem.

NocoBase v2.3.0-beta.8 added an authenticated federation API for AI knowledge base retrieval: AI employees can now pull content from remote, access-controlled knowledge bases with a built-in provider. The authenticated retrieval model closes the gap between internal knowledge governance and AI agent access — the same tension that's driving most enterprise AI deployments right now.

Float shipped AI experience profiles that surface who has done similar work before, directly in the scheduling UI. The tool previously answered "who is available" — now it answers "who has done this kind of work before, and when." Alongside multi-currency rate cards shipping the same week, Float is expanding from scheduling tool to project economics layer.

Wildcards

My Hours shipped as a remote MCP server in Claude's connector directory — no install, no configuration. Claude can log time to specific projects and tasks, retrieve timesheet summaries, and perform time tracking operations conversationally. Time logging is one of the highest-friction recurring tasks in project management; making it a native language operation for anyone using Claude is a practical elimination of that friction.

Themes that compounded

• AI agent architecture rebuilds are arriving: Fibery's complete replacement of its router-plus-sub-agents design with a single dynamic-tool-loading agent is a pattern that will repeat across the sector as first-generation AI integrations hit production limits.
• 1.0 GA with open-source agent stacks: Tracecat open-sourcing presets, skills, and MCP catalog connectors alongside stable release signals a maturity posture — enterprise teams can audit what runs their automations.
• MCP reaching time and project data: My Hours brings time logging into agent context; Tracecat hardens the MCP connectivity layer — both sides of the MCP stack advancing in PM tooling.
• Resource intelligence expanding beyond availability: Float's AI experience profiles add expertise context to scheduling decisions, the kind of information currently stored in manager memory.

Watch this week

Watch Fibery's rebuilt agent in production — the architecture change eliminates sub-agent handoff problems but also concentrates context in a single agent loop, which has its own failure modes at scale. How the team handles edge cases will indicate whether the new architecture is stable or needs another iteration. Also watch Tracecat 1.1.0 alpha — shipping a major alpha the same day as 1.0 stable suggests an accelerated release cadence driven by enterprise pilot feedback.