Fetch.ai’s ASI is extending its Personal AI model beyond single-response assistance through Planner Mode, a system designed to break complex requests into smaller tasks and coordinate their execution. Instead of requiring users to manually choose tools or specialist agents, the planner determines what work is needed, executes each stage and feeds the results back into the plan until it can produce an answer.
The architecture connects ASI with Agentverse, Fetch.ai’s marketplace of independent AI agents. When a task requires capabilities outside the model itself, Planner Mode can search for an appropriate agent, send it instructions and incorporate the resulting output into subsequent steps. Tasks grouped within the same planning round can run in parallel, giving ASI a built-in mechanism for distributing work across multiple specialized services.
Planner Mode Turns User Goals Into Executable Tasks
The workflow begins with a normal request rather than a predefined automation script. ASI decomposes that request into tasks, decides whether each can be handled internally or requires an Agentverse agent, and repeatedly updates its plan as new information becomes available. The product is therefore positioned around goal-driven orchestration rather than forcing users to construct every intermediate step themselves.
Agent discovery can also happen during execution rather than only before a job starts. Fetch.ai’s documentation gives the example of a travel request that might first discover a flight agent and subsequently identify a visa agent based on the information returned. This dynamic discovery allows the execution path to change as specialist agents produce new inputs.
The planner still operates within explicit boundaries. Runs are limited to 15 planning rounds, agent discovery can take up to 90 seconds, and unsuccessful agents are dropped after two failed attempts during the same session. A run can also pause when an agent requests payment, confirmation or additional information. Autonomy is consequently bounded by execution limits and intervention points rather than operating as an unrestricted background process.
Agent Delegation Moves ASI Beyond Conventional Chat
Fetch.ai separately supports user-controlled spending limits for autonomous payments. Personal AI users can allocate dedicated budgets and require confirmations before transactions are completed, with Fetch.ai describing support for payment paths including Visa credentials, USDC and FET. Those financial limits provide a permission layer for transactional agents, but they are separate from Planner Mode’s general task-planning controls.
For developers, ASI’s agentic models can also discover and orchestrate Agentverse services through an API, while the asi1-ultra model supports as many as 500 tool calls within a turn for longer workflows. The infrastructure is being designed for tasks that require repeated searching, evaluation, tool use and specialist-agent coordination rather than a single model response.
The structure does not guarantee that every delegated job will complete successfully. Fetch.ai explicitly notes that independent agents can respond slowly, refuse requests or fail altogether, and some workflows require the user to resume execution after a payment or confirmation request. The more meaningful capability is controlled orchestration: ASI can decompose, route and parallelize work while preserving defined points where execution stops or returns to the user.
That distinction gives Fetch.ai a more concrete agentic proposition than simply adding autonomous language to a chatbot. ASI is being built as an execution layer where user intent can become a multi-stage plan distributed across specialized agents, with the durability of that model ultimately depending on the reliability, cost and usefulness of the Agentverse services it coordinates.