01
Full workflow discovery
We sit with the people doing the work and map the real process — including the undocumented parts. Skipped if you've already done the Agent Opportunity Audit.
← ALL SERVICES
BUILD 01
3–5 weeks
Agent Ops Sprint
Replace the chasing, approvals and copy-paste between systems with agents that run the process — and a single queue where your people handle what the agents shouldn't decide.
- agents
- exception queue
- integrations
- typescript
- next.js
- postgres
WHO THIS IS FOR
Built for you if…
- 01
Teams of 20–200 people where one process — intake, invoicing, scheduling, order handling — eats hours of chasing every week.
- 02
Operators running the business on spreadsheets, inboxes and reminders who want the repetitive part to run itself.
- 03
Companies that tried no-code tools and hit the ceiling — or tried an AI pilot that never left the pilot.
WHAT YOU GET
Everything included.
02
Agents on the repetitive steps
The parts of the workflow that are rules plus judgement — triage, chasing, data entry between systems, first-draft responses — handed to agents with explicit escalation paths. Where a plain script is the right answer, you get a script; we don't bill you for a model you don't need.
03
A human exception queue
Everything an agent won't decide lands in one reviewed queue with the context attached, so nothing falls into the gap between automated and forgotten.
04
Agent boundaries and an audit log
What each agent may do, must escalate, and may never touch — written down and enforced in code. Every autonomous action is logged with its inputs.
05
The tool around it
Where the process needs one: an operations dashboard, lightweight CRM or internal tool, built around your process rather than a template, with role-based access.
06
Integrations where they save time
Email, Slack, calendars, accounting or whatever your process already depends on — connected via API so double entry disappears.
07
Data migration from what you use now
Your existing spreadsheets, exports or legacy database imported and validated, so the tool launches with real data.
08
A before-and-after number
We time the process before we start and measure it again after launch: hours per week, time-to-done, and how often the agents escalate. You get the figures in writing.
09
Deployment + documentation
Hosted, backed up, and documented well enough for a new hire to be useful in an afternoon.
10
2 weeks of post-launch support
The window where real usage finds the rough edges. We smooth them — and tune the escalation rules against real traffic.
WHAT IT LOOKS LIKE
Some of what you get back.
- MAYDraft replies, classify, enrich records, chase overdue steps
- MUST ESCALATERefunds, anything over threshold, unrecognised counterparty
- NEVERMove money, delete records, email customers unreviewed
Every autonomous action logged with its inputs. Anything outside the remit lands in a human queue with context attached.
- cost / run€0.11
ceiling €0.50
- p50 latency740ms
budget 2s
- daily spend€38
cap €80 → queues
- escalation rate6.2%
tuned weekly
Budgets are code, not a monthly surprise. Crossing the per-run ceiling switches to a cheaper path; crossing the daily cap queues and alerts.
HOW IT WORKS
The process.
01
Process mapping + baseline
Days 1–3We walk through the current workflow end to end, time where the hours go, and pick the steps agents should own.
02
Boundaries + core flows
Week 1Agent boundaries, data model and the exception queue, reviewed with your operators before a line of production code.
03
Build + shadow mode
Weeks 2–4You get a working URL early. Agents first run alongside your team in shadow mode, so we can compare their decisions with your people's before they act on their own.
04
Go live and hand over
Final weekAgents switched on step by step, data migrated, accounts created, and a walkthrough with the people who will use it daily.
TIMELINE & PRICING
What it costs.
DURATION
3–5 weeks
BUDGET RANGE
€8k – €30k+
Typically €12k – €18k
PAYMENT
40% to start · 40% at midpoint · 20% on delivery.
WHERE IN THE RANGE YOU LAND
- ·€8k–12k: one process, one or two roles, agents on two or three steps, data that migrates cleanly.
- ·€12k–18k: where most of these land — several roles, an internal tool around the agents, two or three integrations.
- ·€18k–30k: multi-team processes, messy legacy data, or integrations with systems that fight back.
- ·€30k+: company-wide operational change is a programme, not a sprint — see below.
- ·Done the audit first? Its fee is credited in full and discovery is already paid for.
- ·Fixed price once scope is locked, so an overrun is our problem rather than your invoice.
- ·Hosting, model and third-party service costs are billed to your accounts.
BEYOND THE SPRINT
Handing several departments' processes to agents — dozens of users, migration off a system of record — runs as a phased programme. The sprint is deliberately the small version: one process, proven, before anyone commits to that.
STACK
What we build it with.
- Agents
- TypeScript
- Next.js
- Node
- Postgres
- Redis
- Tailwind
NOT INCLUDED
Where the line is.
- ×Replacing your ERP or accounting system of record.
- ×Agents that move money, delete records or contact customers without review.
- ×Native mobile apps — the tool is responsive web unless quoted otherwise.
- ×Ongoing monitoring and tuning after the support window (that is Agent Care).
- ×Open-ended feature requests after the scope is locked — those become the next sprint.
FAQ
Questions we get.
- What does an agent actually own here?
- The repetitive middle of a workflow, with hard boundaries: it never sends money, never deletes, and never decides anything irreversible without a human. Everything it does is logged, and anything outside its remit goes to the exception queue.
- How do we know an agent isn't quietly doing the wrong thing?
- Before going live, agents run in shadow mode next to your team so we can compare decisions. After launch every autonomous action is recorded with its inputs and reasoning, and the exception queue makes silence visible. Any step can be switched back to manual without the tool breaking.
- How do we know it was worth it?
- We measure the process before we start and again after launch, and write both numbers down. If the agents aren't saving time, you'll see that too.
- We already tried Airtable / Notion / Zapier. Why is this different?
- Those are excellent until your process needs judgement, real permissions or real integrations. This is the version that handles the judgement steps, doesn't hit a ceiling, and that you own outright.
- Do we have to abandon our spreadsheets on day one?
- No. We migrate the data, and teams typically run both for a week. Spreadsheets get dropped when people stop opening them, not because we said so.
- Our process isn't documented anywhere. Is that a problem?
- It is the normal case. Process mapping exists precisely because the real workflow lives in people's heads and in three unnamed spreadsheets.
- What happens when we need changes later?
- You own the code and it is a conventional stack any competent developer can pick up. Agent Care covers ongoing tuning; follow-on sprints are scoped the same way as this one.
BUDGET · €8k – €30k+
Which process eats the most hours each week?
Describe it in a few sentences. We'll tell you what handing it to agents would take, and what it should save.
Start this project →