This Workshop Covered REWRITE — The First Phase of ExO 3.0
REWRITE is one of three phases in the broader ExO 3.0 framework — the diagnostic and planning engine that measures organizational shape, identifies binding constraints, and produces the transformation plan.
The full ExO 3.0 journey flows from REWRITE (diagnose + plan) into DRIVE (build the system) and SHAPE (build resilience) — all wrapped around a central Intelligence Stack.
Today's Focus
REWRITE
Phase 1
Diagnose → Understand → ExO 3.0 Plan → Publish
DRIVE
Phase 2
Audit → Design → Build a System
SHAPE
Phase 3
Audit → Design → Build Resilience
The Two Paths to AI-Native
🏢
Direct Mode
50 or fewer employees
- Score and design for the whole company at once
- No immune system strong enough to block transformation
- Brute force the rewrite — move fast, don't wait
- Embed agents across every workflow simultaneously
- Build the minimum viable intelligence stack immediately
AI-Native
Organization
Plan
Resilience
System Design
Intelligence Stack
Measured by
density of intelligence
+ speed of decision
not headcount
🚀
Edge Mode
More than 50 employees
- Score the mothership, then design an edge venture
- Create an AI-native digital twin at the edge
- Start moving workflows over one at a time
- Immune system will attack — expect antibodies
The 7 REWRITE Dimensions — The Organizational Equalizer
1
Organizational Drag
LegacyApprovals, committees, review theater. Internal friction kills speed.
AI-NativeFast execution at lowest competent level. Decisions at the edge.
2
AI Elevation
LegacyIsolated tool. AI lives on the side, not in the core.
AI-NativeCore design principle across the executive layer. Intelligence is central.
3
Work Architecture
LegacyFixed roles. Rigid job titles and waterfall task ownership.
AI-NativeFluid tasks. Dynamic assignment across humans + AI agents.
4
Firm Boundary Design
LegacyAI as external software. Hard walls between humans and machines.
AI-NativeAgents as formal team members with roles and accountability.
5
Decision Autonomy
LegacyEvery decision requires explicit human approval. Bottlenecked.
AI-NativeIntelligent routing between agent-decided and human-decided actions.
6
Network Structure
LegacySlow hub-and-spoke hierarchy. Centralized control chokes throughput.
AI-NativeDistributed authority via a live intelligence network.
7
Reinvention Cadence
LegacyCalcified and fronce. Redesign happens on a decade cycle.
AI-NativeContinuous corporate rebirth. Recursive self-improvement loops.
How You Score It
Score each of the 7 dimensions from 1 to 10 based on how your organization operates today — 1 is legacy, 10 is AI-native. Trust your gut.
/70
Maximum REWRITE Score
7 dimensions × 10 points each
Band-to-On-Ramp Mapping
< 30
Pre-Foundation
Survival risk. Start with education + small MVIS pilots.
30 – 49
Foundational Work
Attack your two lowest dimensions. Run the 90-Day Sprint.
50 – 70
REWRITE Ready
Foundation is set. Proceed to full transformation.
Shape beats total. A balanced 42 structurally outperforms a lopsided 48. Your two lowest dimensions are the binding constraints — they gate everything else.
Pick the Target Workflow + Score Insights
Pick a Workflow — Not a Function
Choose where coordination dominates judgment. Start where a practical first move can be made quickly. Do not pick a department — pick a single, specific, end-to-end workflow.
📄
Invoice Approval
Highly prescriptive: check credentials → legal → finance → payable schedule. Perfect for full agent loop.
👤
Customer Onboarding
Prescriptive intake, categorize, lookup, respond, follow-up. Strong agent readiness candidate.
🎧
Inbound Support Triage
Route, classify, respond, escalate. Intelligent routing between agent-decided and human-decided.
🔬
Sales Research Prep
AI researches company, builds specific outcome format. Frees human for relationship and close.
Score Insights from the Session
Pattern insight
The shape pattern of all 7 scores outweighs the total score. A lopsided 48 means two dimensions are your binding constraints — attack those first.
Don't boil the ocean
A balanced 42 can outperform a lopsided 48. Focus sequentially. Do not chase everything at once or nothing gets done.
Binding constraints
Your two lowest dimensions are your leverage points. They act as drag on everything else. Fix them first to unlock the whole system.
Salim's key point
Context that senior leadership holds is
becoming less useful going forward. Agent-driven environmental scanning replaces gut-level judgment calls.
Task Decomposition Matrix + Deployment On-Ramp
Break the Workflow into Tasks — Score Agent Readiness (1–5)
| # | Task (Example) | Agent Readiness | Score |
| 1 |
Intake / Receive |
|
5 |
| 2 |
Categorize / Classify |
|
4 |
| 3 |
Lookup / Research |
|
3 |
| 4 |
Respond / Act |
|
2 |
| 5 |
Follow-Up / Learn |
|
3 |
5
Agent handles today, no oversight. Deploy immediately.
4
Agent handles with light human review. Near-term.
3
Shared handling / human leads. Build toward agent.
2
Human leads, agent assists. Use as co-pilot now.
1
Fully human: judgment, ethics, relationship. Keep human.
Deployment On-Ramp — Your Three Gears
1
First Move (MVIS)
Target: Weeks
One workflow, one event bus, one agent registry, one agent per class. Do not overbuild. Pick the task with the highest readiness score and run it. Prove the loop.
2
Sprint
Target: 4–12 Weeks
A focused pilot to prove value and harden the agent model. Run the full loop: Sense → Interpret → Decide → Act → Learn. Measure. Iterate. Scale what works.
3
Full Rewrite
Target: 12–24 Months
Complete transformation framework. Deploy, measure, learn, iterate across your edge or direct environment. Create continuous feedback loops between agents. Redesign. Reinvent.
Event Bus Architecture
Two granular agents doing separate tasks outperform one agent doing both. Chain them with a feedback loop on an event bus. Sense–act–learn at each node. Each workflow stage gets its own assistive loop.
Live Coaching from the Session — Real Workflows, Real Advice
Alan
Professional Network / Community
Workflow Tasks
IntakeScreenEvaluateApproveOnboard
Focus on Task 3 (highest readiness). Wrap it in a feedback loop — another set of agents that asks after every run: what can we make better, faster, or simpler? Add one oversight agent above the others. That is your intelligence stack in miniature.
Puneet
Jewelry Design & Manufacturing
Workflow Tasks (7–8 Steps, Only 1 AI-Driven)
SketchDoodleTeam VoteClient ApprovalTechnical DesignVariationsFinal Model
Move variations upstream — assist designers while doodling and sketching to amplify creativity. Wrap the whole waterfall in a collaborative design board (like Google Docs for design). Each layer in the waterfall can have its own assistive loop.
JJ
Enterprise / Senior Management Context
Workflow: Context Flow to Senior Leadership
Project PlanLeadership ReviewDecisionApproval
Build the Knowledge Battery (Kent, Chapter 8) — capture 100% of transcripts, not summaries. Create an internal AI agent that people tap into for organizational context. Add an approval rubric so leaders control what goes in. This is long-term memory for your org.
The 90-Day Transformation Blueprint
1
Target Your Lowest Dimension
Find the binding constraint. Score all 7 dimensions. Your lowest two are your highest leverage.
Days 1–14
2
Select One Workflow
Where coordination dominates judgment. Pick one end-to-end flow — not a function, not a department.
Days 10–21
3
Break into 4–5 Critical Tasks
Map each task. Score agent readiness 1–5. Identify the two highest-readiness tasks to automate first.
Days 14–28
4
Run MVIS Sprint or Full Rewrite
Deploy. Measure. Learn. Iterate. Use sense–interpret–decide–act–learn loop. Show leadership the results.
Days 21–60
5
Execute in Edge Venture or Direct Co.
Run MVIS, Sprint or Full Rewrite across your edge or direct environment. Re-score in 6 months.
Days 60–90+
From the Session
"The firm of the future won't be measured by the size of your workforce. It will be measured by the density of intelligence you have and the speed of decision leaps."
— Salim Ismail
"At my previous company I had a team of 27 people. Now, with AI, I got my team back."
— Workshop participant (founder)
"Don't go to Agile first — leapfrog straight to the agentic AI. By the time you make that transition, it'll be too late. Rip the Band-Aid off quickly."
— Salim Ismail
"Tech equals easy, people equals hard. That's why we want to get rid of the people — I'm just kidding. But yeah."
— Workshop Chat / Salim
"Go, learn, do. Go, learn, do. Some days you're a one or two. Some days you're a five. It ebbs and flows. Keep going."
— Kent Langley
Principles that Compound
🤖
AI-native, not human-native. Design for agent-first, then layer humans in where judgment is irreplaceable.
⛔
Don't automate broken human-to-human workflows. Redesign the workflow first, then automate.
🧱
Build a minimum viable intelligence stack based on business need — not tool excitement. One registry, one event bus, one agent per class.
🔗
Use granular agents with handoffs over an event bus. Two focused agents outperform one generalist agent every time.
🔋
Build the Knowledge Battery. Capture 100% of transcripts, not summaries. Create interfaces to it. Compound organizational intelligence over time.
🔄
Recursive self-improvement at the workflow level. Once agents handle a workflow, build a feedback loop that makes the workflow better each run — without a human in the loop.
📖
Learn, do, learn, do. Nobody has done this before. There are no experts. Move, measure, iterate. The only wrong move is standing still.