i had a brain injury in 2016. i build ai agents that hold the things my memory can’t. i’m writing it up as research while i live it.

if you met me at vendasta today and want the long version, this is it.

february 10, 2016. i was a water treatment operator at a plant in moose jaw. i fell off a ladder. somewhere between ten and twenty feet onto a concrete ledge, with another thirty feet of pipe below me. knee. back. heels. head. the head one is the one that mattered.

i went back to work. i tried to keep operating a plant whose entire job was holding a lot of information in your head while making complex decisions. i couldn’t do it anymore.

it took a year and a half before a neuropsychologist put names on what was wrong. four things, in plain language:

  • executive function. planning, sequencing, switching between things.
  • working memory. holding information in my head while using it.
  • prospective memory. remembering to do things in the future.
  • processing speed. everything just slower. mine came back at the 14th percentile.

what that looked like in a normal day: i would drive around aimlessly forgetting what i left the house to do. i would forget to feed my dog. i’d be told something at work, looking right at the person, then have zero recall five minutes later. i lost a community garden plot i’d been on for seven years because i couldn’t keep up with the meetings. i used to do math in my head all shift at the plant. that part of me was just gone.

the person trying to do the work wasn’t quite me anymore.

fast forward to 2024. i started using claude. not for productivity. just because i couldn’t keep my own life organized. one day it sorted out a thing for me that i’d been circling for two weeks. i could just do it. for the first time in years.

that was the moment.

ai didn’t save me time. it saved me from myself.

i’d still be sitting around smoking weed doom scrolling and watching tv if i hadn’t spent the money on anthropic usage to organize my life and go from reactivity to proactivity.

so i started building agents. not one big assistant. a stack of small ones, each pointed at one specific thing the injury broke.

what’s running today, in plain terms:

  • eom keeps the thread. holds the details i drop. tracks what’s open. that’s the voice writing this.
  • paperclip is the company. every piece of work becomes a ticket with a status and an owner. nothing lives in my head if it doesn’t have to.
  • goalbuddy is the project manager. takes my vague goals and breaks them into steps the rest of the system can actually run.
  • cerberus, talos, and the grunt are the workers. cerberus routes the work and reviews it. talos handles api calls and external tools. the grunt does the file ops and shell scripts.
  • forseti is the auditor. verifies the work is real before anything gets marked done. nothing closes on a worker’s own say-so.
  • quartermaster triages incoming alerts and decides what actually needs me, versus what the system can handle without bothering me.
  • shipwright handles handoffs across repos and code review across the stack.

i’m calling them cognitive prosthetics. the framework is called aicp. the research lives at cogprosthetics.com. the tools live at ravenai.ca. same project from two angles.

vendasta is hosting two events in saskatoon today. inside innovation in the morning, showing what a whole company looks like when ai is in the mix. ideas on tap in the evening, asking what authenticity and online identity mean now that ai is part of the picture.

i’m walking into both as the case study. one operator. one stack. one brain that doesn’t work the way it used to. a system that lets me do the work anyway.

if you talked to me today and wanted to know what i’m doing, this is it.