See what your agents actually accomplish
A new workforce dashboard connects agents, recorded work and human review—so the next conversation is about performance, not how many agents you deployed.
33 posts. Tags are topics; for the editorial split see Feature releases and Developer updates.
A new workforce dashboard connects agents, recorded work and human review—so the next conversation is about performance, not how many agents you deployed.
The software-factory plugin: a request becomes work a person approves from one card, a contract a headless worker executes in its own checkout, evidence QA checks against the contract, and a release the asker hears about. Three pages, four seats, gates at every handoff. vocion-core#655, #658, #672, #683, #695.
Three findings in twenty-four hours — 'approve' bound to the wrong record, a card narrated as a sentence, a card lost in the auto-file round trip — were each a seam between the pieces that make a card. What shipped: the card persisted on the turn and replayed to the model as the tool call it was, written down before it is filed. What is next: one typed card contract in core. vocion-core#690, #694.
Handoff gates: a grammar declared on an object type's transition, a deterministic check that refuses and returns weak work to the seat that produced it, and a judge with the seat's own rubric that passes, returns with a reason code, or escalates to a person. Why the next seat is not the judge, why there is no second review product, and what the first reference runs measured. vocion-core#672, #683.
One room per engagement — a deal, a project — holding everything ingested about it and everything written from it: the sources filed with a weight and a score, the rules the agent reads first, the notes, the timeline, the open items on Review. A collector files clear matches after every sync and asks about plausible ones. What it is, who it is for, and the line that turns it on.
A client document cannot be exported until it has been read as the buyer. One Marketplace with two tabs is where you add capability; Teams is where you see what you already have. The Review queue calls a proposed action what it is — a recommendation. And the data-room collector files things, which it had never once done in production.
Two verbs, two places: Teams and Agents show the capability you already have, Marketplace is where you add more. And a Review item is now a recommendation — because a proposal is a document a client reads, not a decision on your queue.
The screen where you decide what an agent recommended used to open with a restated subject, a confidence number and a paragraph of reasoning, and put the email two thirds of the way down. Now the email is the first thing on the page and renders as a composer, the decision is one sticky bar, and the note beside it changes the verb. vocion-core#466.
A decorative <kbd> showing a shortcut letter sat inside every decision button in the dashboard. Nothing marked it decorative, so it joined the accessible name: the Snooze button was called "Snooze s". A screen reader announced that, and a browser test clicking by exact name never matched — which is how a missing Snooze verb hid behind a failing test for a day. Release notes 2026-09-16 — vocion-core#367.
Vocion's agents now carry typed, scoped memory — preferences, knowledge and procedures, filed per workspace, agent, user, mission or workflow. Every rule is a file a person can read and reword, approval is still the only way one is written, and approved rules finally reach the model instead of waiting to be noticed. Release notes 2026-09-16.
A workspace's voice used to be a paragraph in a prompt that nothing checked. It is now a typed rule set your workspace authors, enforced by a synchronous lint at every seam a draft can come through — the drafting model, the proposal, and the rewrite. A banned phrase costs the model one corrective retry and then fails loudly. And a phrase a reviewer cuts becomes a proposed rule. Release notes 2026-09-16 — vocion-core#370.
Release notes 2026-09-15 — vocion-core#321 (merged): an action proposal carries suggestedDecision (approve, reject, snooze); the review API filters on it; adoption pairs the human's decision with the agent's advice. A reject recommendation can hold a run back from auto-execute, never release one.
Release notes 2026-09-15 — vocion-core#318 (merged): approving or rejecting an agent-proposed action no longer writes a generated rule into the agent's learning step; it queues a candidate for the feedback classifier, deduplicated, for a person to accept.
Every decision a person makes in the review queue is written to a ledger. A six-rung ladder per action kind reads that ledger, and a promotion is refused unless the numbers are there. The thresholds, the schema, and how it maps onto trust.yaml.
Release notes 2026-09-15 — the ask primitive and the Needs-you inbox (vocion-core#325, merged): approvals, rulings, merges and gates as one list with decision sheets, a recommended option, and an always-on Other.
Review and Needs you were two doors onto the same job. They are one now: one list, one detail screen, one action bar, one keyboard map — whatever kind of decision it is. Release notes 2026-09-15 — vocion-core#348.
Release notes 2026-09-15 — a set of review-queue changes on vocion-core main since v2.5x: a failed execution is named and retryable, the queue filters by card type with true counts, Regenerate is an action capability whose feedback is a learning signal, and a run is claimed before it resumes so a double click runs it once.
Every proposal in the Needs-you inbox is now described the way a person would describe it, grouped by the record it concerns, with a real age and columns that line up. Release notes 2026-09-15 — vocion-core#335, merged.
The page where you decide what an agent proposed now says what you are looking at, folds the queue behind you out of the way, and keeps the decision in a bar that never scrolls off. Release notes 2026-09-15 — vocion-core#337.
How to build an AI agent approval workflow you can self-host: an approve step, a skill that requires review, trust thresholds, and a REST decision API.
Which version of the agent did that? A concrete schema for AI agent versioning and audit trails: version rows on apply, a workspace SHA on every tool call.
When a human edits, approves, or rejects a draft, that decision can become a rule the agent reads next time — but only after a person approves it.
An agent can extract and score an application, but only a person should reject one. Here is the object, the gate, and the API call that enforces it.
Vocion's review queue API: GET /api/v1/reviews lists an org's pending agent actions, POST /api/v1/reviews/decide approves, rejects, or edits one.
Build a support agent that summarizes and drafts ticket replies, then holds them in a review queue until a human approves.
A guide to deploying AI agents in production with human approval: five insertion points, an autonomy ladder, trust thresholds, and a REST review queue.
A small team wants shared agents, shared memory, BYOK and an autonomy dial. Here is what assembling it costs, and what one install still lacks.
Autonomy levels are usually an analogy. Here are three real missions at levels 1, 2, and 3, the code that enforces the gate, and what the levels do not do.
v2.26.0 gives every lead a dedicated page at /gtm/lead/{hubspot_id} — the decidable card, the research record, and the decision trail on one addressable URL.
v2.25.0 lands review card contract v2 — subject, provenance, recommendation, typed content and per-object verbs — so any object type presents through one template and decides through one path.
v2.24.0 closes the API gap — fourteen new /api/v1 endpoints give review, feedback and learnings full parity with the dashboard, and admins can now mint API tokens without shell access.
The review queue becomes a focused one-at-a-time flow with editable, steerable actions — and briefings go team-scoped, with a workspace rollup, history, and live regeneration.
Pending work now routes to a specific person — per-person queues, an unassigned triage lane, and snooze — so a workspace with many humans can actually divide the approvals.