Agent Task Manager™ v3.5 is our biggest release since the platform launched. It ships today across all accounts and brings four major surface areas: a rebuilt Analytics tab, an expanded command bar, a new System Health panel, and an AI-powered Recommendations engine. Here’s everything that changed and why each decision was made.
Analytics Tab: Finally, Numbers You Can Act On
The old metrics view showed you raw task counts. Useful, but limited. The new Analytics tab replaces it with a 7-day task completion chart that plots completed, failed, and retried tasks on a rolling timeline. You can see at a glance whether your fleet is accelerating or degrading, and correlate dips with specific deployments or configuration changes.
Below the chart sits the agent leaderboard — a ranked view of your agents sorted by tasks completed, success rate, and average execution time. This sounds simple, but it changes how teams operate. When you can see that Agent-7 is completing 40% more tasks per hour than Agent-3 running the same prompt, you have a concrete starting point for tuning. Clone the top performer’s configuration, apply it to underperformers, watch the fleet equalize.
Performance metrics round out the tab: p50, p90, and p99 latency for task execution; error rate broken down by error type (timeout, model refusal, tool failure); and a throughput gauge that shows current ops/hr against your plan limit. No more guessing whether you’re approaching quota.
Expanded Command Bar
The command bar at the top of every ATM dashboard now supports five commands that cover the most common operator workflows:
- deploy [blueprint-name] — Spins up a new agent from any blueprint in your library with a single line. No UI navigation required.
- task [agent-id] "[instruction]" — Sends an ad-hoc task directly to a specific agent without creating a formal task queue entry. Useful for one-off queries and diagnostics.
- status [agent-id] — Returns the current state, last task, success rate, and uptime of any agent by ID.
- clear done — Flushes completed tasks from the active task list, keeping your dashboard uncluttered during long-running operations.
- help — Lists all available commands with usage examples inline.
Power users who live in the command bar will immediately feel the speedup. We’re also exposing these commands via the ATM API, so they can be triggered programmatically from CI/CD pipelines and external automation tools.
System Health Panel
The new System Health panel appears as a collapsible sidebar on the right edge of the dashboard. It shows four live metrics for your ATM infrastructure: CPU utilization, memory consumption (MEM), network I/O (NET), and operations per second (OPS). Each metric is displayed as a horizontal bar with a numeric readout that updates every 5 seconds.
The bars turn amber at 70% utilization and red at 90%. This gives operators an early warning before agents start throttling. The most common failure mode we saw in v3.0 was users who had no visibility into infrastructure pressure until tasks started failing — the health panel closes that loop.
Recommendations Engine
The most experimental feature in v3.5 is the Recommendations engine. Every 8 seconds, ATM’s analysis layer scans your agent fleet, task history, and system health data and surfaces a prioritized suggestion in the bottom-right corner of the dashboard.
Recommendations fall into three categories: performance optimizations (e.g., “Agent-12 has a 23% retry rate on web scraping tasks — consider switching to a headless browser tool”), resource adjustments (e.g., “MEM utilization has been above 85% for 6 hours — consider scaling your instance”), and workflow improvements (e.g., “3 agents are executing the same search task sequentially — consider parallelizing with a fan-out blueprint”).
Each recommendation includes a one-click action and a dismiss button. Dismissed recommendations are suppressed for 24 hours. We’ll continue refining the signal quality as we collect feedback — the goal is for every recommendation to be immediately actionable, not just observational.
Blueprint Library Integration
Agents can now be deployed directly from the Blueprint Library without leaving the dashboard. Browse your saved blueprints, preview the configuration, and hit Deploy — the new agent appears in your fleet within seconds. Combined with the deploy command bar shortcut, this eliminates the multi-step setup flow that slowed down frequent deployers.
Blueprints also now support environment variable overrides at deploy time, so you can use the same blueprint across staging and production with different credentials and endpoints without maintaining duplicate configurations.
Looking Ahead: ATM v4.0
v3.5 closes the observability gap that users reported most often. The v4.0 roadmap focuses on two areas: natural language task creation (describe what you want a task to do in plain English, and ATM will generate the task definition) and team collaboration features including shared dashboards, role-based access control, and comment threads on individual agent runs. We’re targeting a beta for v4.0 in Q3 2026.
Update your dashboards and let us know what you’re seeing. The telemetry you send back shapes the next release.