Editorial
Research Report
The AI Transformation 100: 100 Concrete Ideas for Fixing How We Work
A few years ago, I sat in a leadership meeting where three senior executives debated a simple question.
"Why did we approve that exception?"
The discount had been granted months earlier. The CRM showed the final price. Finance had signed off. The deal was closed. Yet no one could reconstruct the reasoning. Someone vaguely remembered a service incident. Someone else mentioned a precedent. The VP who approved it had since moved on. We had the data, but we did not have the context.
That moment has stayed with me because it captures a challenge many organizations face as AI agents enter everyday work. We have systems of record for customers, employees and transactions, but we do not have systems of record for how decisions get made. That gap is quickly becoming the bottleneck to AI transformation.
Investors Jaya Gupta and Ashu Garg argued in a recent essay that the next trillion-dollar enterprise opportunity will not come from another CRM or ERP, but from a new layer that captures "decision traces" and connects them into what they call context graphs. Their premise is that agents can access data, but they cannot reliably automate work without understanding how people resolve ambiguity, apply exceptions and exercise judgment in real time.
Thanks to AI, intelligence is no longer scarce, but organizational memory is. As models become widely available and increasingly capable, differentiation shifts away from raw intelligence and toward institutional knowledge. The organizations that capture and operationalize their context will move faster and with fewer errors than those that rely on memory and informal norms.
From Documentation to Context
For years, knowledge management efforts have struggled. The first wave tried to force people to document everything. The second relied on powerful search functionality to find whatever documentation existed. Both ran into friction and faded away.
What changed is that AI has raised the stakes. LLMs reason over language. If something is not written down in a structured, current way, it effectively does not exist to the system.
In my work with clients, I have argued that documentation has become strategic infrastructure. When strategy, decision rights and process precedents are written clearly and kept current, they become "machine-readable judgment" that AI can learn from. Without that foundation, the value of agentic AI is almost non-existent.
Think of this as a knowledge operating system containing artifacts such as a concise strategy codex, a decision register that records context and rationale, tier-1 process libraries, and a shared taxonomy and search layer so people and systems can find what matters. Those artifacts create clarity for humans. They also create raw material for something more powerful; they create traces.
What a Context Graph Really Captures
Gupta and Garg describe context graphs as a living map of how decisions unfold over time. Workplace AI company Glean adds another layer of precision. It models entities like documents and tickets but also actions as first-class data:
- Who approved what?
- Which tools were used?
- In what order?
- With what outcome?
In Glean's words, the system shifts from modeling "what exists" to modeling "how change happens." This nuance is critical because a system of record tells you the current state, but a context graph outlines the path that led there.
That path includes the messy realities of work. The Slack thread that escalated the issue. The informal approval on a call. The workaround everyone uses but no one documented. When these traces accumulate across people and time, patterns emerge. The graph becomes a probabilistic playbook. Agents can learn what typically happens next, when an exception is warranted, and which paths lead to better outcomes.
Over time, the graph compounds. Every human decision adds another example. Every agent run adds another trace and can be evaluated and improved. The result is not a static knowledge base; it is an evolving memory of how the organization actually operates.
The Next Competitive Advantage
For executives leading AI transformation, context graphs represent a new category of asset. For decades, we invested heavily in systems of record because they protected financial and operational integrity. Context graphs protect judgment.
That matters most where work is ambiguous and exception-heavy (e.g., deal desks, incident response, compliance reviews, hiring and promotions). These are the domains where "it depends" is the honest answer and where institutional memory lives in people's heads.
HR leaders, in particular, should pay attention. People decisions are rarely binary. They rely on precedent, values and trade-offs. Without traceable context, AI tools in hiring, performance or compensation risk creating inconsistency and legal exposure. With structured lineage, they become explainable and defensible. The functions that rely most on judgment stand to gain the most from this approach.
The Risks Leaders Must Confront
The caveat here: When you model "how work usually happens," you risk encoding the wrong habits. Patterns can reflect biases, heroics and informal power structures. The fact that a path is common does not mean it is right. If historical traces show that certain employees consistently received faster approvals or more generous exceptions, a naïve agent could reinforce inequity rather than correct it.
Context graphs scale judgment, but they can also scale dysfunction. Consequently, governance cannot be an afterthought. Leaders must decide which signals count as "successful," which behaviors to reward and which to filter out. AI can draft and surface patterns, but accountable owners must approve what becomes precedent. In other words, human oversight remains essential.
Getting Started Without Boiling the Ocean
If the idea of building a context graph sounds ambitious, start with simple, pragmatic steps:
- Document the decisions that matter most. Identify the 50 critical decisions and processes that drive risk and value. For each, record context, options considered, the rationale and the owner. A lightweight decision register goes further than any shiny, new tool.
- Instrument your workflows. Auto-scribe meetings, capture approvals in systems rather than hallway conversations, and treat actions as data. Every structured trace improves future learning.
- Create a shared vocabulary. Agree on what "onboarding complete" or "priority one incident" means. Without a common taxonomy, neither humans nor AI agents can reliably drive business outcomes.
- Make documentation part of performance expectations. Tie relevance, recency and reuse to performance goals. Power should come from what leaders make reusable, not what only they know. Adding to the organization's broader corpus of knowledge should be a part of everyone's responsibilities.
- Pilot one high-value workflow (e.g., incident response, mid-market deal cycles, new-hire onboarding). Capture the steps, measure time and outcomes, and let agents assist within guardrails. Learn and expand from there.
These moves sound mundane — that's the point. Context graphs are built less through grand architecture and more through disciplined habits.
A New Kind of System of Record
Enterprise software's last era captured objects (customers, employees, orders). The next era will capture something more subtle and more valuable — the reasoning that connects those objects.
Webinar
Mar
26
Do Learning Programs Really Work? How to Turn Education Into Engagement In Healthcare
See how leaders are using learning programs to build trust with healthcare professionals and create measurable engagement.
Register
Webinar
On demand
Fuel Your Front Line: Workforce Trends Shaping 2026
Get expert insight into the tools and tactics top frontline teams will need in 2026.
Watch Now
Webinar
On demand
Fix the Disconnect Between Projects and Profit
Your margins are hiding in disconnected data. Let's find them.
Watch Now
Webinar
On demand
AI in Customer Service: Faster Resolutions, Happier Customers
Don’t let rising demand burn out your team. See how to build a smarter, more resilient support org.
Watch Now
When organizations codify how they think, decide and operate, they create an enduring asset that both humans and machines can learn from. Gupta and Garg call that asset a context graph. Others may call it process intelligence or organizational memory. Whatever the label, it is fast becoming foundational.
The leaders who treat it as infrastructure will build companies that learn faster than their competitors. And the next time someone asks why an exception was granted, the answer will not live in memory or lore; it will live in a system that remembers.
Editor's Note: For further reading on the topic, try:
- Knowledge Graphs: The Secret Sauce Behind AI Development — Pairing knowledge graphs with LLMs boosts AI’s reasoning, accuracy and explainability.
- From Tool to Teammate: How AI Is Rewiring People Strategy and What HR Can Do to Adjust — HR leaders see AI transforming work beyond automation — reshaping teams, culture and people strategy. The future is “human-engaged” work, not human-replaced.
- Why AI Readiness Relies on Data Architecture, Fabric and Knowledge Graphs Working Together — AI can’t deliver value without the right data stack. Learn how architecture, data fabric and knowledge graphs create the foundation for reliable, scalable AI.
Learn how you can join our contributor community.