tl;dr: We capture what happened everywhere. But ask 'why did we do that?' six months later and everyone's scrambling. The solution is embarrassingly simple.
There’s a lot of noise right now about “decision traces” and “context graphs.” Fancy words for a simple problem we’ve been ignoring for years.
We don’t need new infrastructure. We’ve just been lazy.
Decision Trace
• Removed feature X
• 2% usage across all users
• 10hrs/week maintenance drain
• Team voted 4-1 to remove
• CEO approved in Monday sync
• 12 affected users migrated
Now you actually know why.
The problem is embarrassingly simple
Every tool we use captures what happened. Activity logs, audit trails, version history. We’re swimming in “what” data.
But ask “why the hell did we do that?” six months later and everyone’s scrambling through Slack threads, pinging people who’ve left the company, reconstructing context from memory.
The data was there. At the moment of decision, someone knew:
- Why this approach won
- What other options got shot down
- Who said yes
- What past decision made this okay
We just didn’t save it. Not because we couldn’t. Because no one asked us to.
We already solved this for code
Engineers figured this out 20 years ago.
Git commit + PR description + linked issue = decision trace for code.
Engineers who’ve been burned treat commit messages like documentation. Not because anyone forces them to. Because they’ve asked “wait, why did we write it this way?” one too many times.
The code doesn’t change. The context around it does.
Now apply that thinking everywhere else
Click cards to flip
Sales
Gave 20% discount
tap to see the why
Sales
Gave 20% discount. Customer had 3 unresolved critical tickets, similar deal got exception last quarter, VP approved citing churn risk
tap to flip back
Design
Changed button color to blue
tap to see the why
Design
Changed to blue. A/B test showed 12% higher clicks, client approved in Figma comment, aligns with new brand guidelines v2
tap to flip back
Hiring
Rejected candidate
tap to see the why
Hiring
Rejected. Strong technical skills but culture fit concerns raised by 2 of 3 interviewers, similar profile didn't work out last quarter
tap to flip back
Product
Removed feature X
tap to see the why
Product
Removed feature X. 2% usage, 10hrs/week maintenance cost, team voted 4-1, CEO approved, migrated 12 affected users to workaround
tap to flip back
Same pattern. Just applied beyond code.
So why hasn’t this happened?
Two reasons.
Friction. In the moment, no one wants to fill out a “reason” field. They’re busy. They’ll do it later. They never do.
Incentives. Activity logs are easy to build. Check a compliance box, ship it. Decision traces take effort. You actually have to think.
What changes now
AI agents fix both problems.
No more friction. The agent is already doing the work. It pulled data from 5 places, weighed options, got approval. Writing down why? That’s just one more line. Zero extra effort.
Incentives finally align. AI agents need past decisions to get smarter. If you never saved why exceptions were made, the agent can’t learn when to make them. Bad data in, bad decisions out.
If you’re building AI agents, you have a choice. Capture the why from day one, or end up with another pile of useless logs.
The moat isn’t the graph. It’s the discipline.
You don’t need fancy infrastructure to start. Just capture three things with every action:
- What happened
- Why
- Who approved
Try it yourself
A text field works. A simple schema works. The value isn’t in the architecture. It’s in having the data at all.
Six months from now, when someone asks “why did we decide this?”, you’ll either have an answer or you won’t.
Want the trillion-dollar insight?