Venture 02

Interactive plans

Plans any agent can write, and I can annotate from any device.

Partial success

The idea

The idea is simple: any agent can create, read, update and delete a plan document, and I can read and update it from any device.

The solution

I built it on my existing AI engineering infrastructure. Each plan is a hosted page I can comment on from my phone, and the agent can ask me questions directly, each with its own answer box and tappable options.

A backend captures every comment, and through MCP any building agent can read the plan, pick up my answers and update it.

The outcome

Partial success.

It does what it was built to do, but it surfaced a limit that wasn't technical: my attention.

19plans written
148notes and answers
12/19plans that got a reply
0notes since 14 Sept
Plans created per week0246827 July: 53 Aug: 710 Aug: 131 Aug: 37 Sept: 3Notes and answers I left per week05010015027 July: 10727 July3 Aug: 393 Aug10 Aug: 110 Aug17 Aug24 Aug31 Aug7 Sept: 17 Sept
Plans created per week0246827 July: 53 Aug: 710 Aug: 131 Aug: 37 Sept: 3Notes and answers I left per week05010015027 July: 10727 July3 Aug: 3910 Aug: 110 Aug24 Aug7 Sept: 17 Sept
Plans kept arriving; my replies didn't. 148 notes and answers across 19 plans, and 7 plans never got one. The last note is from 7 Sept.Source: plan pages' file birth times, and every note and answer synced down from the plans' notes store; weeks start Monday, Melbourne time.

The learnings

What went right

  • Letting the agent ask me questions directly worked. 135 of my 148 notes were answers to its questions, and any agent could pick up the plan and my answers where the last one left off.
  • It works well when I actually read the plans.

What went wrong

  • Plan fatigue. Having detailed plans for everything can be overwhelming, and it leads to fatigue in making the right decisions.
  • I really felt the limit of my attention when working on multiple personal projects.