M.A.I. Consulting
Service 05 of five

Custom Agents, Skills & Tools

“Can the repetitive part be configured once, properly?”

The situation

The same task is done by hand every week, differently by each person.

A grant report drafted from the same template. A call for proposals read and summarised. Meeting notes turned into actions. Each person approaches it their own way, the quality depends on who did it, and the method leaves with whoever wrote it.

The work here is configuration on platforms you already pay for. We do not fine-tune models, build data pipelines or write custom software.

What you hold at the end

A task that is performed the same way every time, by whoever runs it.

Examples

Agents, skills and tools are three different things.

They differ in scope and in what they cost. The definitions below are the ones used in the estimator.

Agent

A role, configured

A configured assistant for a role or workflow, holding your templates, your rules and your reference documents. It handles a multi-step job from beginning to end.

Example: a grant-report drafter that takes field notes and the previous report, follows the funder's headings, flags every figure it could not trace to a source file, and refuses to proceed if beneficiary identifiers are present.

Skill

A method, reusable

A written instruction set the platform loads when a task of that type comes up. It carries your house method, so any staff member gets the same approach without knowing how to ask for it.

Example: a “donor report house style” skill holding your tone rules, your headings, your do-not-say list and your figure-verification step. It applies whether the person is drafting, editing or reviewing.

Tool

One task, repeated

A small, tested set-up for a single repetitive task, with a fixed input and a fixed output. The narrowest of the three and the quickest to configure.

Example: a call-for-proposals summariser that returns eligibility, deadline, budget ceiling and required annexes as a fixed checklist, in the same order every time.

Further examples

Skills worth writing once

A safeguarding language check applied to anything beneficiary-facing. A logframe indicator formatter. A translation reviewer that compares a draft against the source and lists what changed in meaning. A board-paper summariser that always produces decision, options, risk and recommendation.

Further examples

Tools worth configuring once

Meeting notes to actions with owners and dates. A procurement comparison table from three supplier PDFs. A donor due-diligence brief from public sources. A monthly report skeleton pre-filled from the previous month.

What we will not configure. Anything that decides eligibility, safeguarding, hiring or payment without a documented human review. Anything requiring beneficiary data to sit inside a tool with no data processing agreement. If a request falls there, we say so and propose the version that does not.
How we get there

Chosen with you, tested on your documents, handed over with a guide.

  1. Choose the task. Frequency times the time it takes, minus the judgement that has to stay human. We will say when the answer is that it is not worth configuring.
  2. Configure on the platform you already use, at your account level, not ours.
  3. Test against a set of your real documents, including the awkward ones, and record what it got wrong.
  4. Clear against the policy, then hand over with the operating guide, the test set and a named owner.

Typical duration: Two to six weeks

Price

Fixed price and fixed scope, agreed in writing.

Priced per agent, per skill and per tool in the estimator, since each is a discrete piece of work. Configuring them alongside training costs less than configuring them alone, because the teams that will use them are in the room.

Thirty minutes establishes whether this is the right service.

If a different one fits better, or none does yet, we will say so on the call.