👋 Executive Summary
Two weeks ago, I argued that the next $100 billion in enterprise AI won't be spent on models.
This week, we're getting a clearer picture of what enterprises may pay for instead.
Salesforce has introduced Koa, a reasoning model co-engineered with NVIDIA and post-trained from Nemotron specifically for enterprise work. NVIDIA says it draws on 27 years of Salesforce CRM intelligence. Salesforce's rationale is revealing: a general-purpose model approaches a refund policy much like any other reasoning problem. Enterprise work requires something different.
At the same time, enterprise coding-agent company Factory raised $200 million at a $5 billion valuation, roughly tripling its valuation in months.
Put those two signals together and a bigger question emerges:
When AI can generate the work, what exactly are enterprises paying for?
Increasingly, the answer may be what the model doesn't know.
Your proprietary context. Your workflows. Your controls. Your ability to turn intelligence into completed work—and completed work into measurable business outcomes.
When intelligence becomes abundant, value moves downstream.
📋 Today's Docket
🏛️ Executive Brief: Intelligence Is Getting Cheaper. Context Isn't
For the first phase of generative AI, the model was the product.
Better reasoning. Larger context windows. More modalities. Lower inference cost.
That race continues.
But enterprise differentiation is beginning to move somewhere else.
Salesforce's Koa is an important signal because Salesforce did not try to build another frontier model from scratch. It took NVIDIA's Nemotron and post-trained it for enterprise CRM reasoning, embedding accumulated knowledge about how enterprise work actually happens. Koa also operates inside Salesforce's trust boundary.
That architecture points toward a different enterprise AI equation:
The model supplies intelligence.
The enterprise supplies meaning.
As general-purpose intelligence becomes more available, enterprise value migrates toward the layers that understand, execute, govern, and measure proprietary work.
The implication for executives is significant.
The strategic question is no longer only:
Which model should we buy?
It is becoming:
What does our organization know about how work gets done that a general-purpose model cannot know?
📐 AI Executive Framework: The Enterprise AI Value Migration Stack™
Enterprise AI value is moving through six layers.
The closer a layer gets to a measurable business outcome, the harder it becomes to commoditize.
The Enterprise AI Value Migration Stack™ — value moves downstream as general intelligence becomes easier to access.
Layer | What creates value | Defensibility |
|---|---|---|
General Intelligence | Model capability | Declining |
Domain Intelligence | Specialized knowledge | Moderate |
Proprietary Context | Enterprise-specific knowledge | High |
Workflow Integration | Embedded operational position | High |
Governed Execution | Permission + control + accountability | Very high |
Business Outcome | Measurable economic impact | Highest |
This changes the build-vs-buy conversation.
You probably do not need to build a frontier model.
But you do need to decide who owns the layers between that model and your P&L.
The enterprise moat isn't access to intelligence. It's the proprietary system that turns intelligence into outcomes.
📊 Boardroom Debrief: You probably do not need to build a frontier model
But you do need to decide who owns the layers between that model and your P&L.
Ask three questions:
If every competitor had access to the same frontier model tomorrow, what part of our AI system would still be proprietary?
Which enterprise workflows contain institutional knowledge that exists nowhere in our formal data model?
Who owns the execution layer between an AI recommendation and a financial outcome?
Recommended KPI: AI Outcome Yield
Track:
Business value generated ÷ total AI operating cost
Include model consumption, orchestration, integration, human review, remediation, governance, and infrastructure—not just token cost.
Boardroom Decision
Stop measuring AI primarily by access and usage. Start measuring how efficiently intelligence becomes completed, governed work.
The real AI economics problem may sit between inference and outcome—not inside the model itself.
🚀 Startup Spotlight: Factory — Selling Completed Engineering Work
Factory is a useful test case for this thesis.
The enterprise coding-agent company raised $200 million at a $5 billion valuation this week. Reuters reports that its enterprise customers include organizations such as NVIDIA, Blackstone, RBC, Palo Alto Networks, Adobe, and T-Mobile.
The interesting part isn't simply that AI can write code.
Code generation is rapidly becoming available from multiple providers.
The strategic question is what sits around generation.
The Architecture: AI agents operate across engineering workflows rather than functioning only as autocomplete. The value proposition moves from generating snippets toward completing increasingly complex engineering tasks.
The Unit Economics: The enterprise buyer should compare the total cost of AI-assisted engineering against accepted production output—including review, rework, integration, security, and failures. Cheap generated code can still be expensive software.
The Takeaway: Coding agents illustrate the broader value migration. As generation commoditizes, differentiation moves toward workflow integration, reliability, governance, and completed work.
And that brings us back to the question:
When AI writes the software, what exactly are enterprises paying for?
Not merely code.
They are paying for the probability that generated work can safely become production work.
Building an AI startup or enterprise control solution? Submit your platform to be featured in an upcoming edition →
⚙️ The Execution Layer: Map Where Your AI Value Actually Lives
Most organizations can name their AI vendors.
Far fewer can identify which layer of their AI architecture actually creates defensible economic value.
Step-by-Step Implementation
Pick one production AI workflow. Choose something tied to revenue, cost, risk, or cycle time—not an experimental chatbot.
Map all six layers. Identify the model, domain knowledge, proprietary context, workflow integrations, execution controls, and final business outcome.
Run the commoditization test. Assume your competitor gets your exact model tomorrow. Identify what remains differentiated.
================================================================================
THE EXECUTIVE PROMPT STACK: ENTERPRISE AI VALUE MIGRATION AUDITOR
================================================================================
Role:
Act as an enterprise AI strategist and operating-model economist.
Context:
I am evaluating an enterprise AI workflow to determine where its defensible business value actually resides.
Workflow:
[DESCRIBE WORKFLOW]
AI/model providers:
[LIST]
Enterprise systems involved:
[LIST]
Business outcome:
[REVENUE / COST / MARGIN / RISK / CYCLE TIME / OTHER]
Task:
1. Map the workflow across these six layers:
- General Intelligence
- Domain Intelligence
- Proprietary Context
- Workflow Integration
- Governed Execution
- Business Outcome
2. For each layer, identify:
- What we own
- What a vendor owns
- What a competitor could easily replicate
- What would be difficult to replicate
- Where switching costs exist
- Where economic value is captured
3. Run this scenario:
"Tomorrow every competitor receives access to the exact same frontier model."
4. Identify which competitive advantages survive.
5. Identify where value leaks between AI output and measurable business outcome.
Output Format:
A. Value Migration Map
B. Commoditization Risk Table
C. Proprietary Advantage
D. Value Leakage Points
E. Three Executive Actions
F. One KPI to Track
================================================================================Don't use this exercise to justify building proprietary models. Use it to determine which parts of your operating system should remain proprietary.
Save this prompt. Run it against one production AI workflow before your next AI investment review.
📡 Executive Watchlist
Labs — Salesforce + NVIDIA: Koa brings specialized enterprise reasoning into Agentforce, post-trained on Nemotron and Salesforce CRM intelligence.
Funding — Factory: The enterprise coding-agent company raised $200M at a $5B valuation, putting another major valuation marker on autonomous engineering.
M&A — Cohere + Aleph Alpha: The companies signed a definitive agreement to combine in a deal aimed at enterprise and sovereign AI; Reuters reports the combined company is valued at $20B.
Regulation — AI governance: Governments and industry leaders continue debating how much external oversight advanced AI requires as increasingly autonomous systems raise control questions.
Enterprise — Domain models: Salesforce's Koa is another signal that enterprises may increasingly combine general-purpose foundations with specialized reasoning, proprietary context, and controlled execution rather than relying on one general model for every task.
📈 Executive Scorecard
Dimension | Rating | What it means |
|---|---|---|
Model commoditization pressure | High | Access to strong general intelligence is spreading |
Value of proprietary context | Very High | Enterprise-specific knowledge increasingly determines usefulness |
Workflow integration importance | Very High | Intelligence without operational access cannot complete work |
Governance requirement | High | More execution means more permission, accountability, and control |
P&L measurability | Critical | AI spending ultimately needs to survive an outcome test |
Bottom line: The enterprise AI stack is shifting from a model-selection problem toward an operating-system problem.
⚖️ Executive Verdict
The model still matters.
But owning the best model may not be the same thing as owning the best enterprise AI business.
Salesforce's Koa illustrates one direction: take capable general intelligence and specialize it with years of domain knowledge.
Factory illustrates another: move beyond generating artifacts and deeper into completing enterprise workflows.
Both point toward the same economic destination.
Intelligence is becoming an input.
The enterprise value sits increasingly in the proprietary context surrounding it, the workflows connected to it, the controls governing it, and the business outcomes produced by it.
Executive priority: HIGH
💬 Boardroom Question
Ask your CTO, CISO, or AI platform lead:
If every competitor received access to our exact AI models tomorrow, what advantage would we still own?
If every competitor received access to our exact AI models tomorrow, what advantage would we still own?
🛠 AI Tools to Watch
Salesforce Koa — Specialized reasoning for enterprise CRM workflows.
NVIDIA Nemotron — Open models increasingly becoming foundations that enterprises and software companies can specialize for specific domains.
Factory — Enterprise software-engineering agents pushing AI from code generation toward workflow execution.
💬 From My Desk (@DrReemAlattas)
📊 How was today's edition?
Help shape next week's briefing by dropping a comment below.
Forward this to the person governing your AI agents
The AI Executive is for founders and executives who want AI to show up on the P&L—without creating unmanaged operational risk.
The AI Executive is an executive-level publication by Dr. Reem Alattas, focused on AI strategy, enterprise operations, and build-tier execution.
📩 Interested in sponsoring The AI Executive or featuring your startup to our network of enterprise leaders and founders? Contact the partnership team