Learn · Articles
AI-Enhanced Workflow at Manna
I have had a lot of requests about my #AI enhanced process to run Manna Insect company (sales, marketing, budgeting, etc) and how we continuously cultivate our internal knowledge base and guidelines for AI #agents to know what they should be doing.👉 We run a local AI club in Oulu, Finland at Kielo Growth and try to share our practical experiences. So this post is based on those discussions.
01 · The workflow
A practical AI workflow
I am technical person, but beyond coding, it doesn't require that much technical expertise - more a new thinking of HR: when you hire people, you acquire human capacity and intelligence, now you are acquiring digital capacity and intelligence.
Here's a practical simple example of an AI enhanced workflow:
- 1. I ask a fresh agent to get familiar with our marketing strategy and last decisions and plans, and company strategy
- 2. From initial core guidance the agent knows which folders and files to read and then summarises the current plans and next steps
- 3. I ask the agent to create the next post and come back for me to evaluate
- 4. Agent analyses our insights, topic ideas, communication guidelines, previous posts and creates a first draft for review
- 5. I give feedback and ask to proceed and finish the post and accompanying documentation
- 6. Agent finishes the documents and logs all actions and new insight into marketing memory
- 7. I post it and ask to log this work for the next agent
Compared to "human" way, this process saves 45-60 minutes time, but more importantly a lot of my personal mental capacity of getting details right and focusing on the writing. Now this takes about 15 mins, but it takes me 5 mins of mental time (rest is waiting for the agent).
02 · Getting started
Problem #1: how to get started - get an #agentic tool installed locally
This is the biggest question always as most people are still at browser based "chatgpt" level. Unfortunately this will not cut it for long as you always have to drag and drop all files and whatever outputs is provides, you need to reattach and manually move around.
You need an agentic tool installed locally. Previously I would have recommended claude code, codex or opencode CLI tools, but being aware that terminal interface is extremely uncomfortable to most people (please learn it!), now I recommend #Claude desktop app (Cowork) - and at least the $20 plan.
03 · Set up
Start with an empty folder
Then create an empty folder as a starting point and start with a first task, like creating a marketing plan. Ask AI to create a marketing folder and index.md inside it for it to continuously update all plans, decisions and details to it (i.e. memory).
The idea is simple: don't create / edit any files yourself. Ask agent to do it. If it does something wrong, then ask it to update its own guidance. The strategy is to make sure any other fresh agent can get up to speed with the details and continue the work exactly the same.
In the next post I go deeper into memory / guideline examples.
How do you create AI enhanced processes? Do you need these kinds of tips?
Keep exploring
Related topics and pages
- BSF business course — where documented assumptions shape a business case
- BSF start-up course — the validation steps a small team can run with shared records
- Articles library — the written guides this post joins
Read next
- Your AI Agent Has Amnesia. Here's How We Fixed It at Manna. — the companion post on the shared memory that keeps a fresh agent consistent
- Thoughts from our startup and insect experience — Manna's own notes on testing small and standardising a process
- What does Manna Insect actually do? Part 2 — the business validation and mentoring side of the same way of working
- BSF metrics series — the series that puts real operating numbers behind a BSF case
Next step
Put the workflow to work
Manna can help organise the material, decisions and records behind a BSF case so the next step rests on a shared, written context.