- The Shift Brief
- Posts
- From Context Graphs to Context Failures
From Context Graphs to Context Failures
The Shift Brief | Week of Jan 19th
Ryan Erickson
January 21, 2026
Why Context Is the Difference Between Useful AI and Risky AI in Investment Firms
Last week, we broke down what a Context Graph is and why it matters for investment firms.
This week, a recent Forbes piece sharpened the conversation by calling out a growing realization across enterprise AI: most AI failures are not model failures. They are context failures.
That distinction matters more in investment firms than almost anywhere else.
AI hallucinations are a symptom, not the disease
When people talk about AI hallucinations, the conversation usually stays technical: model accuracy, prompt quality, guardrails.
But hallucinations rarely come from randomness. They come from AI being asked to reason without the institutional context humans take for granted.
In an investment firm, that context includes:
Prior views on a company or sector
Historical decision rationale
Management credibility over time
Internal debates that never made it into a formal memo
Why a position exists, not just what it is
When that context is missing, AI can produce answers that are technically correct but strategically wrong.
Why is this problem amplified in investment research
Investment firms do not suffer from a lack of data. They suffer from fragmented meaning.
Consider a few common scenarios:
An AI tool summarizes an earnings call accurately, but ignores how management missed guidance the last three quarters
A research assistant answers a question without knowing the firm's existing exposure or house view
A compliance team cannot trace how an AI-generated insight was formed or what inputs it relied on
None of these are model issues. They are context gaps.
Context cannot be reconstructed on demand
One of the biggest misconceptions in enterprise AI is that context can be added at query time.
In practice, the institutional context is continuously created through meetings, research notes, internal conversations, and decisions that compound over the years.
If that information is not captured, enriched, and connected as it happens, it cannot be reliably recreated later. Especially under time pressure.
This is why prompt engineering alone breaks down in real investment workflows. You cannot prompt your way to institutional memory.
Context engineering is becoming infrastructure
The Forbes article points toward a broader shift that investment firms should pay attention to: the next phase of AI advantage will not come from smarter models. It will come from firms that treat context as infrastructure rather than exhaust.
That means:
Capturing firm activity as it happens
Preserving decision history, not just outcomes
Making context reusable across people, teams, and time
Ensuring AI outputs can be traced back to inputs and rationale
This is not about adding another tool. It is about building the connective tissue that makes AI trustworthy inside complex organizations.
The takeaway
AI does not fail in investment firms because it lacks intelligence. It fails because it lacks memory.
Firms that invest in context now will not just get better answers; they will also gain a competitive advantage. They will get safer, more consistent, and more institutional outcomes from AI over time.
~Ryan Erickson, Founding Executive