Fibery rebuilt its AI agent from scratch while Atarim made agents' accumulated memory legible and editable — the sector's AI layer is maturing past demos into configurable infrastructure.

The week in project-management

Two releases this week surfaced a tension in how teams are building AI into project management: Fibery rebuilt its agent architecture entirely — replacing a router/sub-agent system with a single streaming instance that the team found easier to improve — while Atarim made its agents' accumulated memory visible and writable by users. The first is a behind-the-scenes architectural bet. The second is a transparency bet: if AI agents hold institutional knowledge about your clients and projects, you should be able to read and correct it.

Tracecat continued building security workflow automation with semantic search and agent session infrastructure, positioning itself as the AI-native alternative to traditional SOAR platforms.

Leaders

Fibery completely replaced its AI agent architecture this week. The old multi-tier router that delegated across specialized sub-agents is gone, replaced by a single streaming instance the team reports is faster to iterate on and easier to debug. OAuth Applications also shipped, enabling MCP connections to external services. Whiteboards gained AI integration, and Space templates with file and image support rounded out the release. Fibery rebuilding the agent layer from scratch rather than patching the old one is a signal that the first architecture couldn't be improved incrementally.

Atarim shipped two releases this week that together define a new capability: AI agents whose institutional memory is visible and editable. Version 5.1.2 makes the memory file that specialists accumulate about each client and workspace readable and writable — teams can see exactly what the agents learned and correct it when it's wrong. The 5.1 release that preceded it cut AI action costs by 40% and added a full SEO research-to-execution loop. Both releases together position Atarim's AI agents as auditable tools rather than black boxes.

Tracecat shipped pgvector-backed semantic search for case management tables in 1.1.0-alpha.2, alongside workspace sync export and pull via MCP. For a security automation platform moving toward agentic workflows, semantic search over case data is load-bearing: agents querying similar past incidents need retrieval that goes beyond keyword matching. The MCP workspace sync also lets external agents pull Tracecat workspace state as a first-class operation.

Wildcards

SmartSuite added EU data region selection at workspace creation — a compliance milestone that opens the platform to organizations that couldn't adopt it under data residency requirements. For a work management platform competing in the European mid-market, this removes the most common procurement blocker. It shipped without announcement fanfare, which suggests it was a demand-driven feature that the team has been building toward for a while.

Themes that compounded

• AI agent transparency is becoming a product feature: Atarim's readable/writable agent memory and Fibery's architectural rebuild both reflect teams' frustration with AI layers they can't inspect or correct.
• EU data residency shipped at SmartSuite this week — a demand-driven compliance feature that reflects how many mid-market European teams the platform has been unable to onboard.
• Agent session backends and semantic search both landed at Tracecat together, reflecting the infrastructure requirements for agents that retrieve context rather than just execute predefined flows.

Watch this week

Fibery's agent architecture rebuild is the kind of move teams make when they have a clear thesis about what to build next and the current architecture can't support it. Watch Fibery's September and October release notes for what the new agent layer enables that the old one couldn't — that will reveal the roadmap faster than any announcement.