AI Should Learn How You Work
When we first designed Snappy PM, a natural idea came up: since we have AI, why not let it read a company's historical projects and automatically generate a workflow that belongs to that company? It sounds reasonable — and very AI-native. But as we kept building, we changed our minds. What happened in the past doesn't mean it should happen in the future. AI that diligently learns history may just end up automating yesterday.
When we first designed Snappy PM, a very natural idea came up:
Since we have AI, why not let AI read a company's historical projects and automatically generate a Workflow that belongs to that company?
It sounds reasonable.
And very AI-native.
But as we kept building, we changed our minds.
Because AI can easily make a simple thing look very clever.
It looks at a company's past projects and can summarize many Stages, Rules, Roles, and Exceptions.
But the problem is:
What happened in the past does not mean it should happen in the future.
A company's own historical processes may already contain many accidental, inefficient, or even wrong practices.
If AI learns this history very diligently, the end result may simply be:
Automating yesterday.
So the new version of Snappy PM made a very important simplification.
We no longer let AI invent a State Machine for every company.
The basic skeleton of a project is five Phases:
Requirement Capturing
Scoping & Design
Implementation or Manufacturing
Deployment or Installation
Closing
This part stays stable.
What AI should really learn is the meaningful differences between companies.
For example:
Which Phases apply to you?
What kind of Quotations do you usually use?
What are your Payment Terms?
Which Documents are your own standard templates?
Do you have Inventory?
Does the customer face this company, or another Sales Company inside the group?
These are:
How you work.
So Snappy can read a past project, understand the company's working habits and documents, and make suggestions.
But we gave it another principle:
Evidence or default.
If AI believes a company works in a certain way, it should be able to tell us:
which real documents I saw this in.
If there is no evidence, use the simple, stable platform defaults — instead of letting AI guess.
I increasingly feel this is a very important boundary in AI product design.
The most valuable thing about AI is not necessarily:
letting AI decide everything.
It is knowing:
what should stay stable, and what can be learned.
Accounting rules should stay stable.
The core Project Flow should stay stable.
Permission boundaries should stay stable.
While a company's documents, habits, templates, and specific ways of working can be learned.
So now we prefer to describe Snappy PM this way:
AI learns how you work.
It doesn't invent how you should work.
AI should reduce the cost of a business adapting to software.
Not create a new layer of complexity and then demand that the business adapt to the AI.
SnapLedger, your life easier.
An important boundary in AI product design: know what should stay stable and what can be learned. Accounting rules, the core project flow, and permission boundaries should stay stable; a company's documents, habits, templates, and specific ways of working can be learned. And when AI claims a company works a certain way, it should be able to show the real documents it learned that from — evidence or default.
If an AI learned your company's way of working purely from your past projects, which of your habits would it faithfully reproduce — and which ones would you be embarrassed to see automated?
AI learns how you work. It doesn't invent how you should work.
Get the Founder Diary + regulatory updates for your country
One email per week, only when there’s something real — Richard’s founder diary and the regulatory updates that matter where you live. Free, unsubscribe any time.