Custom agents
An agent that follows your process
Some jobs are too specific for a general assistant. They follow your procedure, use your templates and touch your systems. We sit with the people who do the work, then build a Bwat agent that takes the same steps and stops for a person wherever it matters.
This is a project we build with you, not a do-it-yourself agent builder.
- 1
A tender arrives by email
Bwat picks it up and reads the documents
- 2
Checked against your criteria
Using your scoring rubric and past bids
- 3
Response drafted
In your house format, with gaps marked
- 4
Manager approves
Nothing is sent until a person signs off
Needs a person - 5
Sent and filed
Saved to the right folder, with a record of who approved it
We start with the job
Before any AI is involved, we map the process with the people who do it: what starts it, what they look at, where judgement comes in, who approves, and what the finished result looks like. Only then do we decide what the agent does, what stays with a person, and what's better as plain software.
- We watch the current process and collect real examples
- We agree what the agent may decide and what it must hand to a person
- We connect only the systems you approve
- We agree how to measure quality before the pilot starts
Good first jobs
Read applications, claims or bids against your own rubric and flag the ones a person should look at first.
From first meeting to daily use
The order matters. It stops an impressive demo from being mistaken for something you can rely on.
- 01
Define the job
One bounded process, the people it serves, and the result worth improving.
- 02
Build on real cases
Write down the procedure, connect the approved material and tools, and test on real examples.
- 03
Pilot it carefully
Run it beside the current process, record what goes wrong, and keep a person at every step with consequences.
- 04
Keep it current
Update the instructions, references and connections as your process changes.
What you get
The process, written down
Versioned instructions for the steps, decisions, templates, exceptions and hand-offs.
Connections you approve
Links to the internal or outside systems the job needs, built and reviewed for your setup.
Clear permissions
Written rules for what the agent may read, change or send, and what it may only suggest.
Test cases
Typical and awkward examples we use to check that each change makes the work better.
The right place to use it
Web chat, VS Code, the Windows app, or an interface built for the job.
Upkeep
We keep the agent in line as your documents, systems and rules change.
Before we scope it
- Isn't this just a saved prompt?
- A prompt is one small part. A custom agent also has reference material, permissions, connections to your systems, approval points and tests for the whole job.
- Can the agent make final decisions?
- That depends on what's at stake and what you approve. We start with people making the consequential calls and the irreversible moves, and automate more only when the results justify it.
- Do we need Bwat Enterprise?
- Yes. A custom agent needs scoping, review, support and a contract, which is what Enterprise covers.
Bring us one process
Tell us what starts it, what information it uses, who checks the result, and what makes it slow or unreliable today.