A small, measured pilot, then your call
We would rather prove value on one task than promise everything. Here is exactly what happens, what we need and what you get.
From first call to decision
- 20 minutes
Intro call
You describe the task, the tools and who does it today. We say whether an agent, a skill or neither is a sensible fit.
- Week 1
Agree the metric
We write down what success means (for example, hours saved or documents received on time), take baseline numbers, agree how each run will be scored and set the scope.
- Weeks 1 to 3
Build
We build the agent or skill on your tools with your wording. Your reviewer tests drafts weekly and we adjust.
- Weeks 3 to 4
Run the pilot
The agent works on real tasks with a person approving what is customer-facing. We score and log every run, and review the misses.
- End of week 4
Decide
We share the numbers against the agreed metric, with the run scores and what did not work. You choose whether to continue.
Your side of the pilot
- One person who knows the workflow and can answer questions quickly.
- A reviewer who approves drafts, ideally for about 30 minutes a day during the pilot.
- Sample data: exports, documents or message examples, with any needed consent.
- Access to the tools involved, limited to what the workflow needs.
- Agreement on the success metric and the baseline.
At the end of four weeks
- A working agent or skill on one task, running on real work.
- A short report: the metric, the baseline, the result, the run scores and the issues found.
- Documentation of how it works and what it connects to.
- A clear recommendation, including "do not continue" if that is the honest answer.
A fixed-fee pilot first
The pilot has one price, agreed in writing before we start. If you want to continue, we then agree a monthly plan based on what the agent does and how much it is used.
- Pilot: fixed fee, typically 4 weeks. We agree the fee in writing once the work is scoped, before we start.
- Monthly plan: agreed after the pilot. Optional. Includes support and maintenance.
- No long lock-in is required to start.
Engineering and QA teams, and broking and trading firms: scope and fee are agreed per engagement, because the work varies more. See engineering and QA and capital markets.
Typically 4 weeks, with a success metric agreed upfront and human approval on anything customer-facing. Monthly plan agreed after pilot.
Book a 20-minute callWe keep it running
An agent is not a one-time build. Messages, forms, tools and business rules change, and agents need looking after. If you continue after the pilot, support and maintenance are part of the optional monthly plan.
Monitoring
We watch for failed runs, errors and odd results, and tell you what we find.
Fixes
When a tool changes or something breaks, we repair it. Questions and issues go to one contact.
Updates
New wording, rules, templates or products are updated as your business changes.
Regular reporting
A short monthly summary against the metric we agreed at the start, including issues.
Quality checks
Runs keep being scored. When scores slip, we update the agent or skill and test the fix before it goes live.
Room to grow
When one workflow is stable, we can add the next one on the same terms.
Scope, response times and the monthly fee are agreed in writing before the plan starts. The plan is optional and you can end it with notice, as set out in that agreement.
Engineering habits behind every agent
Agents fail in boring ways. These are the rules we build and test against.
A checklist, not a feeling
Every requirement goes on a list. An agent is not finished when a demo works, only when each item is built and checked.
Blockers are written down
If something is missing, it is recorded as a blocker with the reason, never skipped quietly. Everything not blocked still gets done.
A second opinion on key calls
Important judgements, such as why a test failed, can be reviewed by a second AI model before a person sees them. No result is called a pass without evidence.
Run. Score. Review. Improve.
Every run is scored against checks you agree on. A person reviews the misses, the agent or skill is updated and tested, and the score trend shows whether it is really getting better.
No secrets in prompts or logs
Credentials stay in your environment settings. Agents get the access they need and no more.
Autonomy is earned
Agents start by drafting, and your people approve everything. As the scores prove it, routine cases can run on their own, while anything sensitive still goes to a person. Nothing irreversible happens without approval.
Careful by default
Minimal data
We ask for the data the workflow needs and nothing more, with access limited and removable.
Human approval
Customer-facing messages are approved by your people. Agents do not file, pay or place orders on their own.
Handled in line with the DPDP Act
Personal data is handled in line with India's Digital Personal Data Protection Act, 2023. We do not hold security certifications at present, and we will not imply otherwise.
You own your data. We do not sell it. Where we need to delete data after a pilot, we do so when you ask. Read the Privacy Policy for how this website handles your details.
Integration targets
These are the tools we aim to connect to. They are targets, not partnerships: we are not affiliated with or endorsed by any of these providers, and we confirm what is feasible for your setup in the first call.
Not sure where to start?
Tell us about one task that eats your team's time. We will tell you honestly whether an agent can help.