Automation & AI work companion
AI work companion
You've probably tried ChatGPT or Claude already – and forgotten it again after three weeks. Not because the tools are bad. But because an empty chat window doesn't know what your work looks like.
AI work companion
01AI work companion
A work companion turns that around: instead of explaining who you are and what you need from scratch for every task, the AI tool is properly taught your work once – your templates, your tone, your typical tasks, your knowledge. From that come fixed workflows: the inquiry that wants sorting and answering; the report due every week; the research that always follows the same shape. You put the case in, the companion delivers the draft – tailored, not off the shelf. The fastest entry point in the whole catalog: Tier 1 stands within days.
Straight up: a work companion isn't an autonomous employee – it prepares, you decide. It's only built for real, recurring work (tool training without a use case is wasted time). And what it may do with your data is settled beforehand – client data doesn't drift casually into external systems.
02Demo
Three workflows, as they run in set-up companions – from three different businesses. For each workflow, look at what goes in and what comes 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.
03How it works
- 01
Choose the one work area
Which recurring work should get easier? A clearly bounded area beats the do-it-all – the rest comes later, once the first area holds.
- 02
Gather tasks and knowledge
What does the companion need to work along? Templates, examples, rules, tone. In doing so it becomes visible where this knowledge sits today – in heads, emails, documents – and what has to be structured first.
- 03
Build the workflows
From tasks and knowledge come fixed processes: instructions, prompt templates, knowledge files. Not a box of prompt slips, but a set-up tool.
- 04
Anchor it where the work happens
The companion moves into your existing environment – as a set-up Claude/GPT project or later as its own small interface. Without a system break.
- 05
Bring the team along and sharpen
Who uses the companion for what, what works, what's missing? Instruction and persuasion are explicitly part of it – the next stage grows from real use.
04Stages
The fastest quick win in the whole catalog: Tier 1 stands within days, not weeks – and each tier builds on the previous one.
Entry: the set-up AI project
A Claude/GPT project with curated instructions plus an introduction – usable immediately in everyday work, without new software.
Extended: knowledge, skills, team
Knowledge files, a library of fixed workflows for recurring tasks, team training – the companion becomes a team tool.
Full build: your own interface
A small tool of your own with tool integration and automations for the key processes – when Tier 2 hits its limits.
05Technical implementation
- Entry without infrastructure of your own: a custom-GPT or Claude project with curated instructions and knowledge files.
- The knowledge base feeds on your documents: templates, quote modules, standards – as Markdown, Notion or Drive storage.
- Advanced: a small web interface of your own with a backend, model access and tool integration.
- Automations for recurring handovers run via n8n, Make, Zapier or your own 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 rights are checked per use case – not switched on wholesale.
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.