Internal Automations
AI that runs the repetitive busywork
About Centauri Media
Centauri Media is the same digital agency behind The CRM. As it grew, so did the volume of small, repetitive tasks: summarising, drafting, categorising, and moving data between tools, individually trivial but collectively a full-time job.
Objectives
- Take the repetitive layer of agency work off people's plates.
- Use LLMs as reasoning steps inside real business pipelines, not as a novelty.
- Read from and write to the systems the team already uses.
- Tune prompts and guardrails for reliability, so the output is dependable.
The Challenge
A growing agency generates endless small tasks: summarising, drafting, categorising, moving data between tools. Individually trivial, collectively a full-time job that eats the time meant for client work.
The Solution
The automations use LLMs as reasoning steps inside real pipelines, reading from and writing to the systems the team already uses, with prompts and guardrails tuned for reliability rather than novelty. Drafting and summarising happen at the point of work, and data is moved and enriched between systems on its own.
Each automation is a pipeline between the tools the team already uses: read from one system, let an LLM be the reasoning step (draft, summarise, categorise, decide), then write to another, with guardrails around the output so a wrong answer is caught before it ships rather than after. They run unattended against the same data the CRM owns, so the automations and the product share one source of truth.
Repetitive knowledge work everywhere
A growing agency generates endless small tasks: summarising, drafting, categorising, moving data between tools. Individually trivial, collectively a full-time job.
LLMs wired into the pipes
The automations use LLMs as reasoning steps inside real business pipelines, reading from and writing to the systems the team already uses, with prompts and guardrails tuned for reliability rather than novelty.
- Drafting & summarising at the point of work
- Data moved and enriched between systems
- Guardrails so output is dependable
An LLM writing into real systems can't be trusted blindly, so the same guardrails that make the output dependable also cap how much you let it do on its own. The automation is reliable because it's kept on a short leash, not because the model is; loosen the leash and you trade dependability for reach.
Technologies
Conclusion
The repetitive layer of agency work now runs unattended, so the team's time goes to work that actually needs a human, and a small team delivers 25+ projects.
Have a project in mind? Let's talk