Automation & AI work companion
AI work companion
You've probably tried ChatGPT or Claude already – and forgotten about it three weeks later. That's rarely the tools' fault. An empty chat window just doesn't know what your work looks like.
You'll get …
- 1
An AI tool that knows your work
Templates, tone, typical tasks and knowledge are set up once – nobody has to start from scratch with every task.
- 2
Set workflows for recurring work
Inquiries, reports, research: the case goes in, a tailored draft comes out – you check it and decide.
- 3
A tool where you already work
A ready-configured Claude or GPT project in your own setup, with no new software – plus an introduction for the team.
01Intro
A work companion turns this around. Instead of explaining who you are and what you need all over again for every task, you teach the AI tool your work once – your templates, your tone, your typical tasks, your knowledge. You hand it the case, it hands you a draft – made to measure, not off the peg. It's the quickest way in across the whole catalog: Tier 1 is up and running within days.
A work companion isn't an autonomous employee – it prepares, you decide. We only build it for real, recurring work (training people on a tool without a use case is wasted time). And what it may do with your data is agreed in advance – client data doesn't casually end up in third-party systems.
02How it works
- 01
Pick one area of work
Which recurring work should get easier? A clearly defined area beats a jack of all trades – the rest can follow once the first area is working.
- 02
Gather tasks and knowledge
What does the companion need to pitch in? Templates, examples, rules, tone. Along the way it becomes clear where this knowledge lives today – in people's heads, emails, documents – and what needs structuring first.
- 03
Build the workflows
Tasks and knowledge are turned into set workflows: instructions, prompt templates, knowledge files. Not a shoebox of loose prompts, but a properly configured tool.
- 04
Embed it where the work happens
The companion moves into the setup you already have – as a configured Claude/GPT project, or later as a small interface of its own. No switching between systems.
- 05
Bring the team on board and refine
Who uses the companion for what? What works, what's missing? Showing people how it works and winning them over is part of the job – the next stage grows out of real use.
03Stages
Tier 1 is up and running in days, not weeks. Each further tier builds on it.
Starter: the configured AI project
A Claude/GPT project with carefully written instructions, plus an introduction – ready to use day to day, with no new software.
Next step: knowledge, skills, team
Knowledge files, a library of set workflows for recurring tasks and team training – the companion becomes a tool for the whole team.
Full build: your own interface
A small tool of your own, connected to your other tools, with automations for the key processes – for when Tier 2 reaches its limits.
04Try it yourself
Three workflows from three businesses. Give the same case to a blank chat window first, then to the set-up companion – and decide at the end what goes out.
This is how a companion comes about: your real work – templates, examples, rules – is distilled into instructions and knowledge files. That sets up the AI project that delivers drafts. And what you correct day to day flows back as fine-tuning.
The dashed loop is the difference from a box of prompt notes: the companion gets better through use – tier by tier.
05Technical implementation
- Getting started needs no infrastructure of your own: a custom GPT or Claude project with carefully written instructions and knowledge files.
- The knowledge base draws on your documents: templates, quote modules, standards – stored as Markdown, in Notion or in Drive.
- Advanced: a small web interface of your own with a backend, model access and connections to your tools.
- Automations for recurring handovers run via n8n, Make, Zapier or custom scripts.
- Hosting: none needed for the AI project; for an interface of your own, e.g. Vercel/Node/Supabase.
- Data access, client data and tool permissions are checked for each use case – not switched on across the board.
for context · This shows how something like this usually gets built – not how it has to be built for you. What actually fits is worked out in the project: around your existing setup, your team, and what stays easy to maintain.