PolyAI Wren for improving agents; Think² for coordinating delivery review
PolyAI Wren helps prepare agent changes, run simulations, and analyze conversations. Its documentation separates approving changes in a Branch from publishing that Branch to Live: the user normally publishes. One exception is ending an experiment with a winning Branch, which merges it onto Live after prior approval of that action. If you need to bring together the response owner, test acceptance owner, and person authorizing the next action, consider a Think² group with those people and collaborators to prepare a review record with results and pending questions. This is neither an equivalent Wren replacement nor automatic PolyAI integration.
Waitlist: no guarantee of immediate access or a free plan. The destination is in Spanish.
One passed test does not make three tests complete
Fictional exercise, not a PolyAI execution: change R has three planned tests, P1 passed, P2 running, and P3 queued. The package includes approval to apply R to a Branch, but no evidence of publication to Live or agreed acceptance criterion. A useful delivery separates completed work, pending work, and the decision still needed. This establishes neither that R fails, that Wren publishes incomplete changes, nor that all three tests must pass under a universal rule.
Do you need PolyAI changes and simulations or delivery-acceptance coordination?
Keep change R's scope, each test's status and source, and any agreed acceptance criterion. Do not count running or queued as passed, or invent a criterion when none is supplied. Approval to write to a Branch establishes neither functional acceptance nor publication to Live. If tests are not linked to a revision or environment in the package, leave that relationship pending for someone able to confirm it.
Bring content, test, and next-decision owners together
Bring content, test, and next-decision owners together
Think² groups bring relevant people and collaborators together with shared context. Consider this space for preparing a common record showing each owner which results are available and what remains undecided. Reading the record establishes neither test acceptance, task responsibility, nor permission to publish an agent.
Prepare a review without promising a PolyAI test runner
Each collaborator can have its own computer, and compatible models can be selected. The proposal uses authorized artifacts and actual access; it guarantees neither PolyAI access, automatic simulation execution, Branch reading, nor continuous test-state capture. The proposed delivery is a reviewable record, not certified technical validation.
Make the specific action explicit without extending earlier permission
An external action requires specific prior approval of the action, content, and recipients. Permission to send a record does not authorize publishing a Branch, and approval does not establish execution or functional acceptance. A different proposed action must make its scope and available evidence clear before execution.
Waitlist: no guarantee of immediate access or a free plan. The destination is in Spanish.
When to keep your current tool
Keep PolyAI Wren if you want to prepare changes, simulations, and analysis within the PolyAI agent environment. Its documentation supports inspecting, adjusting, or reverting a change before publication and leaves normal publishing under user control; experiment actions also require prior approval. This is not an approval deficiency for Think² to fix. Consider Think² for coordinating reviewers and preparing an evidence record, not as a replacement for PolyAI simulations, Branch management, deployment, or voice service.
A check before switching
Proposed evaluation, not performed: review the fictional change R package and its three tests, record each status and source, separate Branch-write approval from publication evidence, and formulate the pending acceptance-criterion question. PolyAI was not accessed, no Branch was changed, and no simulation was run. Accuracy, safety, timings, and Think²-versus-Wren improvements were not measured.
How to evaluate a deliverable
- Define one delivery with sources, responsible people, tools, and permissions. Do not migrate everything or assume universal compatibility.
- Keep change R's scope, each test's status and source, and any agreed acceptance criterion. Do not count running or queued as passed, or invent a criterion when none is supplied. Approval to write to a Branch establishes neither functional acceptance nor publication to Live. If tests are not linked to a revision or environment in the package, leave that relationship pending for someone able to confirm it.
- Keep PolyAI Wren if you want to prepare changes, simulations, and analysis within the PolyAI agent environment. Its documentation supports inspecting, adjusting, or reverting a change before publication and leaves normal publishing under user control; experiment actions also require prior approval. This is not an approval deficiency for Think² to fix. Consider Think² for coordinating reviewers and preparing an evidence record, not as a replacement for PolyAI simulations, Branch management, deployment, or voice service.
- Change-review record, fictional exercise: revision R / source and scope / Branch and environment if supplied / P1 status and source / P2 status and source / P3 status and source / established or pending test-to-revision link / agreed or unsupplied acceptance criterion / owner able to confirm it / recorded approval scope / applied write if supplied / publication and evidence if supplied / pending question / proposed next action and specific approval / verified execution. Passed, running, and queued are not equivalent states. This record neither runs simulations nor integrates Think² with PolyAI and establishes no production result.
Frequently asked questions
Does approving a Branch change prove publication and that every test passed?
No. The Wren introduction distinguishes change approval from normal publishing by the user. Ending an experiment with the winning Branch merges it onto Live after approval of that specific action. In the fictional exercise, P1 passed does not change P2 or P3's pending states or establish an acceptance criterion absent from the package. The example is not attributed to a real Wren execution. Approving a change also does not mean approving each action an agent might later execute.
Must I replace what already works?
Keep PolyAI Wren if you want to prepare changes, simulations, and analysis within the PolyAI agent environment. Its documentation supports inspecting, adjusting, or reverting a change before publication and leaves normal publishing under user control; experiment actions also require prior approval. This is not an approval deficiency for Think² to fix. Consider Think² for coordinating reviewers and preparing an evidence record, not as a replacement for PolyAI simulations, Branch management, deployment, or voice service.
Is there a demonstrated winner?
Proposed evaluation, not performed: review the fictional change R package and its three tests, record each status and source, separate Branch-write approval from publication evidence, and formulate the pending acceptance-criterion question. PolyAI was not accessed, no Branch was changed, and no simulation was run. Accuracy, safety, timings, and Think²-versus-Wren improvements were not measured.
Sources and public walkthroughs
Keep exploring
Waitlist: no guarantee of immediate access or a free plan. The destination is in Spanish.