Tool

AI Readiness Assessment

Answer 5 questions about a specific workflow to see whether it's a better fit for a chat assistant, a bounded AI agent, or a hybrid of both.

1. Can you list the specific actions the system would need to take?

2. Is the task mostly about answering questions, or about changing something in another system?

3. Is the relevant data or context already accessible in a structured, retrievable form?

4. How many steps does a typical task involve?

5. Do you have a way to grant scoped, revocable permissions to the systems this would touch?

The three pattern-fit results, explained

The tool above scores your five answers and matches them to one of these three patterns. Here's what each one means and recommends, regardless of which one applies to your workflow.

Start with a chat assistant grounded in your own documents

Your answers describe a task that's mostly about informing rather than acting, without a clearly enumerated action set yet. That's the right starting point for a knowledge assistant grounded in your own documents (RAG), not a bounded agent — it's cheaper to build, easier to govern, and often the tool people actually need for this kind of task.

Start by grounding a chat assistant in the specific documents or data this task actually needs.
Revisit the bounded-agent option once the specific actions involved can be listed concretely.
If scoped access control doesn't exist yet for the systems involved, that's worth building before considering an agent that can act, regardless of which pattern you choose first.

A hybrid pilot is the more honest starting point

Your answers put this workflow between the two patterns — some informing, some acting, with data and permissions partially in place. Rather than committing fully to a bounded agent or a pure chat assistant, a scoped pilot on one well-understood slice of the workflow will tell you more than committing to either architecture upfront.

Pick the single most repetitive, best-understood sub-task within this workflow to pilot first.
Build out scoped, revocable permissions for that sub-task specifically before widening scope.
Keep a human review step in the loop for this pilot, even if the long-term goal is a more autonomous agent.

This workflow is a reasonable fit for a bounded agent

Your answers describe a task with clearly enumerable actions, real state-changing steps, structured data access, and multi-step tool use — the pattern where a bounded agent architecture, with explicit tool permissions and approval gates on anything irreversible, tends to fit well.

Design tiered permissions before writing any agent logic: read-only lookups, then approval-gated write actions.
Add human approval gates specifically on irreversible or high-stakes actions, not just a general review step.
Log every tool call and the reasoning behind it, so decisions can be reconstructed after the fact.
Still start with a narrow, well-defined action set — an agent's scope is easier to widen later than to walk back.