After an incredibly successful Beta period, Agentic Tooling 2.0 is now the default and recommended execution mode for all Operator agents powered by reasoning models. The graduation from Beta comes with confidence — user feedback and real-world usage patterns have shaped a system that delivers genuinely agentic AI capable of planning, reasoning, and executing complex multi-step tasks.
Agentic Tooling 2.0 was built for problems that cannot be solved in a single prompt. Research projects, multi-source report generation, software development tasks, and compliance workflows all require exploration, iteration, and verification. AT 2.0 gives agents the structure and tools to work through these problems autonomously while staying accountable.
What Makes AT 2.0 Different
Unlike legacy agentic tooling that relied on fixed upfront plans and sequential execution, AT 2.0 introduces adaptive planning that revises on the fly, parallel execution for independent tasks, and mid-task memory saves that preserve critical context during long-running sessions.
- Adaptive Planning — The agent revises its plan when new discoveries change the picture, reducing wasted tool calls on dead-end paths
- Parallel Execution — Independent tasks run simultaneously, reducing response times on complex requests
- Intelligent Delegation — Sub-agents (Researcher, Analyst, Verifier) work in parallel with richer context for higher-quality results
- Active Skills — Agents actively choose to activate skills like Verify Output, Step Back and Rethink, and Structure Response when the situation calls for it
- Execution Presets — Control how long an agent works: Short (30 min), Medium (1 hour), Long (2 hours), or Marathon (4 hours)
A Team of Specialists
Your agent can now hand off tasks to specialist sub-agents that work in parallel. The Researcher searches your knowledge base, the internet, and deep-dives into sources. The Analyst processes data, runs code, analyses files, and generates reports. The Verifier cross-references sources, performs QA, and validates results. This delegation happens invisibly — you get the result, not a handoff notification.
Specialist sub-agents communicate through shared memory, so each one benefits from the others' findings. If the Researcher discovers a conflicting source, the Verifier is notified and can flag the discrepancy. If the Analyst produces a calculation, the Verifier checks the methodology. This collaboration model produces outputs that are more reliable than single-agent approaches.
Observability and Control
Despite the increased autonomy, you remain in control. AT 2.0 exposes a reasoning trail showing what the agent planned, what it discovered, and why it changed course. You can set execution budgets, require human approval for specific tools, and pause or stop a run at any time.
Standard RAG agents remain fully supported as a beloved modality for many use cases. AT 2.0 is available for all subscriptions and can be enabled from Settings > Agents > [Your Agent] > Agentic mode.
