Skip to content

Agents & Custom Skills

Agents on your team.

Agents that watch, read and produce, and skills that teach AI tools how your team works.

Solutions · Applied AI

Agents do work that runs on its own: watching, reading, drafting and producing, every day. Custom skills put your team’s knowledge and standards into the AI tools it already uses.

We design and build both, and we’re careful about where each one belongs.

Two different things

“Agent” gets used for almost everything now. Agents and skills solve different problems, and many organizations need skills first.

  • Agents

    Agents act on their own.

    They’re triggered by a schedule or an event, do a job with several steps, and report back. A monitoring agent reads its sources every morning. A reporting agent drafts the weekly summary before anyone’s in. Nobody has to start them.

  • Skills

    Skills teach your AI tools how your team works.

    A skill is packaged knowledge and instructions: your proposal process, your writing standards, your pricing rules, your checklist for reviewing a contract. When someone asks Claude to review a contract, the skill means it works through your checklist instead of a generic one.

Work nobody has time for

Agents suit the jobs that never get done because they’d take too many hours. A good monitoring agent doesn’t hand you a feed. It tells you what changed and why it might matter to you, and links the evidence so you can check it.

  1. Watching the field

    Reading competitor announcements, regulatory updates and the publications you follow, and flagging the few that matter.

  2. Hearing every customer

    Going through every support ticket, sales call transcript and survey response to see what keeps coming up.

  3. Keeping the site true

    Checking a large website every week for broken links, outdated facts or pages that have drifted from policy.

  4. Content from a live feed

    Producing fresh versions of content from a live data feed, every day, on a schedule no team could keep.

The judgment lives in the skills

What it knows

Much of the value in an AI tool comes from what it knows about your work. A general model writes a generic proposal. The same model with your pricing logic, your past work, your voice and your checklist writes one your team only has to edit.

Written down

That knowledge usually sits in a few people’s heads or is scattered across old documents. Writing it down carefully, in a form AI tools can use, is worth doing now. It keeps its value as the models underneath change, and it’s the part of the system that makes the output yours. We work this way ourselves: our team uses a library of custom Claude skills we built.

Connected

We build custom skills and plugins for tools like Claude, and connect them to your data and systems through standard integrations such as MCP. Through a system’s API, a skill can look up a client record, check a price list or create a task, as well as write.

When an agent is worth building

Not every task should be handed to an agent. We look for three things.

  • The work repeats.Something done daily or weekly is a far better candidate than something done once a year.
  • The inputs are available.The agent can reach the sources, data and systems it needs, with permission.
  • Mistakes are easy to catch.Where a wrong answer is cheap and visible, an agent can work alone.
When the stakes are high, the review step matters as much as the agent. We design it carefully: what the person sees, what they approve, and how quickly they can correct something.

Built so you can see what it did

An agent that works in the dark is a liability.

Logging

We log what an agent reads, decides and does, at the level the work calls for, so you can check it.

Limits

It can reach only the data and systems it was given, and change only what it’s allowed to.

An owner

Someone on your team is responsible for it and gets told first when it goes wrong.

A pause

When the work calls for it, we build in a way for its owner to stop it while something gets checked.

Agents inside products

Sometimes the agent is the product. When a service depends on reading, synthesizing or acting on information at a scale people can’t manage, such as weekly check-ins from every employee or questions from every customer, the agent sits at the centre of what’s being sold. One prototype we built for a media company interviews an advertiser and turns the conversation into a draft sponsored article. That work lives under New Digital Products, with the same care over review, limits and logging.

See New Digital Products

Once it’s running

  1. Work that finally gets done

    The reading, watching and checking nobody had hours for could happen every day, with the few things that matter flagged.

  2. Less rote work, more judgment

    People could spend less time on repetitive tasks and more on the calls only they can make.

  3. Know-how that stays

    What lived in a few people’s heads becomes something the whole team, and its AI tools, can draw on.

  4. Answers about what it did

    Where the work calls for it, there’s a record of what the agent did and why.

Related

The expertise behind it

All expertise