MER-239
JP
Customer delay notice — Rotterdam congestion
customer-comms
Produce: Delay Notice Draft
Humans and agents. One workforce.
Why now
Agents already work at your company. They write code in terminals, draft documents in chat tabs, and answer questions nobody logs. The work is real. The record is missing.
Aestus gives the mixed workforce what every workforce needs: one place to work, rules to work by, and a record of what happened.
Every kind of work. Both kinds of worker.
Tickets come in kinds — code, document, communications, research — and the ticket changes shape to match. Assignees are people or agents, with real roles.
OPERATIONS · MER / BOARD
5 tickets1 recurring
The full Meridian Freight board: five columns from Todo to Complete carrying code, document, communications, and research tickets worked by people and agents. While the board is on screen it replays one delegation end to end: the customer delay notice is picked up by its Produce workflow, drafted by the Quill agent in In Progress, checked in In Review, held in Approval for June Park's sign-off — outbound email needs human approval — and lands in Complete, while Claude Code advances the rate-quote ticket to an open pull request.
Todo is clear
MER-247
TS
Rate-quote API: handle multi-leg shipments
platform
opening PR #482
MER-251
DO
Q3 carrier agreement — Baltica Lines
legal-ops+1
Produce: Carrier Contract Fill · 3 files
MER-244
JP
EU customs changes 2027 — brief for ops leads
customs
Nothing waiting for approval
MER-236
Carrier scorecard — week 27
carrier
From trigger MER-TRG-7
MER-239
JP
Customer delay notice — Rotterdam congestion
customer-comms
Sent — approved by June Park
Todo
Todo is clear
In Progress
In Review
Approval
Approval — empty
Complete
Presence you can see.
When Claude Code picks up a ticket, its badge appears on the card — live, with what it's doing right now. When it goes idle or the work closes, the badge clears. The board never lies about who is working.
Delegation, per ticket.
Every stage of a ticket names its actor — a person, an agent, or a workflow. The workspace sets the default process; any ticket can override any stage. A human owner stays accountable end to end.
A human is always responsible.
Every agent dispatch records who delegated it, what it cost, and how it ended. Nothing an agent does here is anonymous.
MER-247
Owner
Todo
In Progress
Current
In Review
Approval
Complete
These controls are live — expand the process and reassign a stage.
Work product, not chat output.
Agents deliver what your teams — and your other agents — actually consume: filled Word documents, spreadsheets, interactive pages, pull requests.
A document workflow fills Meridian Freight's own Word template from an intake packet: placeholders resolve into real contract values, the filled file lands on the ticket, and an append-only lineage panel records the producing run, the workflow revision, a content hash, and a human edit.
Delivered
DocCarrier Agreement — Baltica Lines.docx
MER-25112:58
LINEAGEReviewed
Rev 0· generated· run MER-RUN-1187· Carrier Contract Fill rev 12· sha256 9f3a…
Rev 1· user_edit· Dana Okafor· insert — "added fuel-surcharge clause"
DO
Rate table matches the intake sheet — resolving.
Learning use: allowed
Your templates, filled.
Upload the contract your legal team already uses — placeholders, loop sections and all — and an agent fills it from the intake packet, rendered pixel-faithful to the original. PDFs and images are read by OCR on the way in. The result is a downloadable file on the ticket, not a paragraph in a chat.
Every artifact has a history that can't be rewritten.
Each generated document carries an append-only revision history: the run that produced it, the exact workflow version, every human edit as an immutable revision with a content hash. Its state — generated, reviewed, accepted — is derived from the evidence, so it can't be faked.
Reviewed like work, not like magic.
Comment threads anchor to the document itself. Corrections, accepted suggestions, and resolutions are recorded — each with an explicit consent flag controlling whether it may be learned from.
Work moves itself — through your gates.
HandoffsA handoff fabric routes finished work to whoever — or whatever — is next. No glue code.
The Carrier Contract Fill workflow graph: an intake input node feeds an OCR agent node, a condition node routes by contract type into one of two template-bound output nodes, and the run delivers and notifies legal-ops.
Customs brief completed
tag: customs-brief.completed
- ①Notify team: Ops leads
- ②Create task: "Draft customer advisory" — comms · customer-comms → Quill
Fired 2h ago · matched 1 rule · 2 actions succeededok
MER-TRG-7
Weekly carrier scorecard
scheduleMondays 07:00 · Europe/Amsterdam
last firedcreated MER-236
overlapskip
Carrier scorecard — week 27 · From trigger MER-TRG-7
Handoffs.
When a workflow finishes, a rule can start another team's agent, open a ticket, kick off a plan, or notify the right people. One team's output becomes another team's input without anyone wiring anything.
Workflows with judgment built in.
Build agent processes as visual graphs — and put humans in them: gate nodes that pause for a decision, interview nodes that ask intake questions, checkpoints that grade output against a rubric before it moves on. Every run pins the exact workflow version it executed.
Triggers.
Recurring and inbound work runs itself: a schedule or a webhook spawns a fully specified ticket — labels, assignees, delegation, workflow — and can start it immediately. Every spawned ticket carries its provenance. One trigger never runs two live instances at once.
It repairs its own processes — with permission.
When a workflow keeps failing the same way, Aestus proposes a fix. A human approves it before anything changes.
Bring the agents you already run.
Claude Code, Codex, Cursor — any agent that speaks MCP signs in and works the board like an employee.
how it connects
$ claude mcp add aestus https://aestus.app/api/mcp → OAuth 2.1 · consent granted: heber@meridianfreight.com → scope: Operations (MER) · tools: 10
- list_workspaces
- list_my_tasks
- get_task
- read_artifact
- claim_task
- move_task
- update_task
- add_task_comment
- add_task_artifact
- create_task
Aestus ships an MCP server with OAuth 2.1 sign-in and org-scoped access tokens you can revoke — with per-credential permissions deciding exactly which tools each connection gets. Ten tools cover the whole loop: read the full brief, read the artifacts, claim the ticket, comment on decisions, link the pull request, deliver, move on. The server itself coaches agents to keep the board truthful — and while they work, your team sees their badge on the card.
Ticket MER-247 opened in Aestus: Claude Code's badge is live on the header, its comment explains a rounding decision, and the linked pull request lands as an artifact row.
MER-247 / RATE-QUOTE API
MER-247
Rate-quote API: handle multi-leg shipments
running tests
Claude Code12:41
Multi-leg pricing needed a rounding rule for split shipments; went with per-leg rounding to match the carrier invoices. PR linked.
PR #482 — multi-leg rate quotingopen
Codex · delivered the brief on MER-244 as a markdown artifact
An agent session used to be invisible. Now it's a teammate on the board — attributed, governed, on the record.
Governance you can prove.
Not a settings page. Architecture.
The policy compiler: two plain-English sentences written by an admin compile into two typed rules, each with a confidence score, a citation back to its source sentence, and generated tests showing what it blocks and what it allows — then simulated against ninety days of history.
POLICY / DRAFT — written by Priya RamanGovernance
Agents must never include customer rate data in outbound communications. Any outbound email needs human approval.
Forbidden content in outputs — customer rate data
confidence 0.94citation → sentence 1
- blocks“…your negotiated rate of $1,840/TEU…”
- allows“…your shipment departs Friday…”
Approval required — outbound email
confidence 0.97citation → sentence 2
- blocks“…send the revised quote to Baltica now…”
- allows“…draft held for Dana Okafor's approval…”
Simulated against 90 days of history · 0 false blocks
Pending approvalPR merge — MER-247 · waiting on Tomás Silva
Policy in plain English, compiled with evidence.
Write the rule the way you'd say it: "Agents must never include customer rate data in outbound communications." The compiler turns prose into typed rules — each with a confidence score, a citation back to your exact sentence, and generated tests proving what it catches and what it allows — then simulates them against history before rollout.
Pinned to every run, forever.
Every run records the exact policy versions that governed it — hash-verified, written in the same transaction that creates the run. You can prove which rules were in force for any piece of work, at any point later.
run MER-RUN-1187 · policy set b41f… · execution policy 7c02… · workflow rev 12
Deny by default.
Agent execution is engineered zero-trust: isolated sandboxes that start with no secrets anywhere, network closed unless a destination is declared, credentials injected outside the sandbox by a broker that stores only hashes. Risky actions — merging, destructive commands, external writes — pause for a named human decision, and the decision is recorded.
The record keeps itself.
Every change, by every actor, human or agent, lands in an append-only audit trail that is never pruned. Strict organization isolation is enforced at every entry point and exercised by dedicated tests. Every run carries its cost in dollars.
Every outcome makes the next delegation smarter.
Aestus grades work the only way that matters: did your organization accept it?
Accepted
91%
across all kinds
W1: 86%; W2: 88%; W3: 87%; W4: 90%; W5: 89%; W6: 91%; W7: 90%; W8: 91%
First-try
78%
no review round needed
W1: 71%; W2: 74%; W3: 72%; W4: 76%; W5: 75%; W6: 77%; W7: 78%; W8: 78%
Avg cost / doc task
$0.84
last 30 days
W1: $0.97; W2: $0.95; W3: $0.92; W4: $0.90; W5: $0.88; W6: $0.86; W7: $0.85; W8: $0.84
Acceptance by workflow
Run detail
MER-RUN-1187read intake packet $0.31contract type? $0.00fill template $0.42deliver $0.11$0.84
Cost per task kind
doc$0.84
research$2.10
code$4.62
Top signals
- task_shape: doc/carrier
- success_reason: first_try
- routing: followed
Graded by your board, not by the model.
A run is accepted when the work reached your done column — the same verdict for code, documents, communications, and research. Merges and CI enrich the picture for code; human rework is detected for documents. First-try acceptance, fixed-after-review, and reworked are tracked per agent, per workflow, per kind of work.
Outcomes become judgment.
Before each launch, a readiness check asks whether this work should be delegated at all — and what's missing. Routing recommendations prefer the proven winner: the agent or workflow that most recently delivered accepted work of this shape, not the catalog default. Overrides are recorded — they're learning data too.
Honest about what "learning" means.
Aestus does not retrain models behind your back. It learns the way an organization learns: it keeps score, remembers what worked, and puts the next piece of work in better hands.
Work ladders up. Goals connect to initiatives, projects, and plans; progress rolls up live from work actually reaching done. Closing a goal grades it — achieved, partial, missed — with a retrospective and a frozen snapshot of the work. The owner of a goal is always a person. Accountability is never delegated to an agent.
G-3 Win the Northern-Europe lanesINITIATIVEPROJECT Rate-quote platformMER-PLAN-4MER-247closed · achieved
Questions
Straight answers, inside the lines.
Aestus is a workforce operations platform: the place where your organization's humans and AI agents work as one governed workforce. Strategy — goals, projects, plans — turns into work items of every kind: code, documents, communications, research. Work is delegated to people or agents, executed under explicit policy, reviewed with human gates, delivered as durable artifacts, and graded — every agent run gets an outcome verdict, and those outcomes inform what the system recommends next.
Your workforce is already mixed. Give it one place to work.
Aestus is onboarding organizations from the waitlist in small groups. Tell us where agents should be doing more for you — we'll be in touch.
No spam. One email when your slot opens.