AI BRAIN ARCHITECTURE
Your AI stops guessing when it has a brain
We build a knowledge system in your git repo. Your AI reads it before every task. Your output stops being generic.
The problem
Same models.
Different results.
In the GTM teams we audit, most say AI is not landing. The models work. The context around them does not.
Every Monday starts over. Someone opens ChatGPT. Types “write me a cold email.” Gets a stranger’s voice. Spends 30 minutes fixing it. Closes the tab. Next week, same thing. No memory. No judgment. Your 100th prompt is as dumb as your 1st.
The mechanism
Permanent memory for your AI
Markdown files in a git repo. Your AI reads them before every task. Knowledge stays. Chats do not.
Context files
Five files. Identity, offering, team, motion, evidence. Your AI reads them before every task.
Tribal knowledge, written down
Positioning, ICPs, battlecards, signals, personas. Each claim tagged confirmed, inferred, or hypothesis.
Output archive and feedback
Every output saved. Campaign results sync back. The system learns. Output 100 beats output 1.
Three tier skill system
From capabilities to complete plays
Skills stack. Small capabilities become full plays that match your sales motion.
TIER 1
Atoms
- One task. One output.
- Research an account from a domain.
- Draft an email in your voice.
- Score a lead against your ICP.
TIER 2
Molecules
- Atoms chained in order.
- Research, then outreach.
- Signal, then response.
- Loss, then battlecard update.
TIER 3
Compounds
- Full plays with human checkpoints.
- Launch a new logo campaign.
- Run a quarterly refresh.
- Run a competitive displacement.
What you get
Nine deliverables. Full ownership.
Everything lives in your git repo. You own it.
Context Intake System
Five files that hold what your AI needs to know.
Identity, offering, team, sales motion, evidence. Your AI reads them before any task.
Derived Knowledge Files
Your tribal knowledge, written down.
positioning.md, icp.md, battlecards.md, signals.md, personas, sales motion. Every claim tagged confirmed, inferred, or hypothesis.
Three Tier Skill Library
Atoms. Molecules. Compounds.
Single capabilities chain into composite skills. Composite skills chain into full plays. Each skill names the files it needs.
AGENTS.md Resolver
Routing for every skill.
One file maps each skill to the context it needs, what feeds it, and what it produces. No skill runs without the right context loaded.
Signal Library
Buying signals with decay and scoring.
Each signal has a shelf life, a score, and rules for how it combines with others. A job post alone scores differently than a job post plus a tech install plus a funding round.
Sync Scripts
CRM and campaign data flow back into the brain.
Automated sync from your CRM and outbound tools. Results feed the archive. The system learns from real numbers.
Output Archive
Every output saved. Every result tracked.
The archive closes the loop. A reply teaches the next email. A win teaches the next battlecard.
Operating Rhythm
Weekly updates. Retros. Quarterly refresh.
A written cadence so the brain stays current. Not a one time build that rots. Under two hours per week.
Change Control Hooks
Core files protected from casual edits.
Git hooks flag changes to foundational files like positioning and ICP. Context loading tiers stop you burning tokens on simple tasks.
IN PRACTICE
The brain changes how you work
Patterns we see when teams build the brain. Not testimonials. Same outcome, same setup.
SIGNAL DISCOVERY
The ICP that rewrote itself
targeted personasPlatform Engineering teamsA dev tooling company found Platform Engineering teams through signal analysis. A persona they never targeted became their fastest growing segment.
BATTLECARD UPDATE
Losses that teach
objection cost the dealnext rep had the counterAfter a loss debrief, the battlecard updated with the objection that cost the deal. The next rep had the counter ready.
ACCOUNT RESEARCH
Mornings back
45 min per accountunder 5 minA cybersecurity SDR team automated account research. Same quality. 45 minutes became under 5.
ONBOARDING
Day one context
week one rampday one contextA RevOps leader joined with full positioning, ICP, competitors, and active plays on day one. All in files she could read.
The difference
AI Brain vs. raw ChatGPT
Same model. Different output.
Cold email quality
Without AI Brain
Generic. Any company.With AI Brain
Your positioning, ICP, battlecards.Account research
Without AI Brain
Starts from zero each time.With AI Brain
Builds on past research and signals.New hire ramp
Without AI Brain
Weeks of shadowing. Slack digs.With AI Brain
Day one access to written knowledge.Competitive response
Without AI Brain
Scramble after the call.With AI Brain
Battlecard with confidence tags.Campaign learning
Without AI Brain
Anecdotal. Lost in Slack.With AI Brain
Saved outputs with results.Signal detection
Without AI Brain
Manual and patchy.With AI Brain
Automated with decay and scoring.Context for AI tasks
Without AI Brain
Copy paste from random docs.With AI Brain
Files load automatically.Knowledge over time
Without AI Brain
None. Every Monday resets.With AI Brain
Each output makes the next better.The model is not the bottleneck. The context is.
How we build it
We build it. You own it.
Build phase: 4 to 6 weeks. Then ongoing rhythm.
- 1Wk 1 → 2
Context intake
Foundation
- Five file intake.
- Stakeholder interviews.
- Audit of existing collateral.
- CRM and tool access setup.
Deliverable
Raw context files in your repo
- 2Wk 2 → 4
Knowledge build
Build
- Derive positioning, ICP, battlecards, signals.
- Build the three tier skill library.
- Create the AGENTS.md resolver.
- Tag every claim with confidence.
Deliverable
Full AI Brain in your git repo
- 3Wk 4 → 6
Activation
Launch
- Sync scripts wired to your tools.
- Output archive set up.
- Operating rhythm written down.
- Change control hooks installed.
Deliverable
Live system with feedback loops
Maintain and grow
- Weekly knowledge updates.
- Campaign retros.
- Quarterly full refresh.
- New skills as needs change.
Cadence
The brain gets smarter every week.
Is this for you
Built for GTM teams burned by AI
Not for everyone. Here is who gets the most out of it.
This is for you
- GTM teams of 3 to 30.
- Founder led or operator led.
- ACV $20k to $500k. Multiple personas.
- Someone will open a folder and run a CLI.
- Past "will AI help" and into "why is the output bad".
- Tribal knowledge is your edge.
This is not for you
- Enterprise with 100+ sellers. Different scale.
- Solo founders pre product market fit. Build the knowledge first.
- Transactional ecommerce. Sale is not complex enough.
- Teams who want to "try AI" with no process.
- Anyone expecting magic from a prompt template.
FAQ
Common questions
Markdown files in a git repo. They hold your GTM knowledge: positioning, ICPs, battlecards, signals, personas, sales motion. Your AI reads them before every task.
One person on the team needs to open a folder, read markdown, and run a CLI. This is not a no code dashboard. It lives in your real workflow.
Prompts are single use. The brain is permanent. Every prompt references the brain. So every prompt has your ICP, positioning, competitors, and past results loaded.
The operating rhythm handles that. Weekly updates. Campaign retros. Quarterly refresh. Change control hooks protect core files. The system is built to evolve.
Any AI tool that accepts context. ChatGPT, Claude, Cursor, custom agents. The brain is plain markdown in a git repo. Tool agnostic.
Most teams feel it in week one. The first cold email that sounds like a teammate wrote it makes the value obvious. It builds from there.