GenAI and the Missing Layer of Organizational Awareness
Several leading industry reports suggest that most Gen AI implementations are failing.
MIT researchers found — in a widely cited, and debated, study — that roughly 95% of enterprise GenAI pilots have yet to demonstrate measurable business value.¹ Gartner predicted in 2024 that at least 30% of generative AI projects would be abandoned after proof of concept; by its own later accounting, the reality passed 50%.² And S&P Global reported that the share of companies abandoning most of their AI initiatives jumped from 17% to 42% in a single year.³
Sticker shock. Underwhelming results. Quiet abandonment.
Here’s my read of these findings — not as someone studying AI from the outside, but as someone who has spent the last few years building with it every day.
I keep coming back to two observations.
First, GenAI performs reliably when its context is bounded by a finite corpus of information with sound provenance. Change any of those variables — let the corpus sprawl, let the provenance become uncertain — and reliability starts to decline. Notably, MIT’s own diagnosis points the same way: the failures trace not to model quality but to a “learning gap” — tools that never learn the organization’s actual context.¹
Second, even with a perfectly curated corpus, GenAI still falls well short of humans at solving abstract problems. It can synthesize, connect ideas, and enumerate possibilities — but where the authenticity or reliability of its underlying training knowledge is uncertain, its findings shouldn’t stand on their own. They need dependable validation checkpoints, with humans retaining the final judgment.
Taken together, these two observations explain much of what we’re seeing. Organizations that began with problems GenAI is not particularly well suited to solve — or that never invested in curating the context it depends on — were always more likely to see disappointment than durable value.
That doesn’t read to me as an indictment of the technology. It reads as a question of fit.
If it were merely the growing pains of an immature technology, the next wave should be faring better. It isn’t: Gartner already expects over 40% of agentic AI projects to be canceled by the end of 2027 — for the same reasons.⁴ The pattern isn’t fading with each new generation of the technology. It’s migrating.
Building and operating AI is expensive. Naturally, organizations look for return on investment, whether through new revenue or lower costs. Workforce reduction has therefore become one of the most visible applications.
Some of that is inevitable. Where AI can reliably perform work as well as — or better than — people, we should be honest about that. But I think the broader framing of GenAI as workforce replacement often points the technology at precisely the kinds of problems where its limitations become most apparent.
Meanwhile, there is another problem hiding in plain sight.
As organizations grow, they accumulate complexity. Plans evolve. Teams specialize. Assumptions drift. Decisions made in one part of the organization quietly affect another. Enormous investments go into management, reporting, and tooling simply to keep everyone moving in the same direction. And despite all of that effort, staying aligned only becomes harder.
One of my guiding principles for GenAI has been remarkably simple: illuminate the dark corners of my awareness, so I can make better decisions. Not replace my judgment — extend my sight.
That principle keeps bringing me back to the same question. What if those dark corners aren’t somewhere out in the world? What if they’re inside our own organizations?
Read the two observations again, but this time in reverse.
GenAI needs a bounded corpus with reliable provenance. An organization is one of the few places where creating that kind of corpus is actually possible. Its plans, discussions, decisions, reports, and artifacts already belong to a shared body of knowledge. Unlike the public internet, that knowledge can be curated, governed, and continuously improved.
GenAI shouldn’t be asked to replace human judgment on abstract questions. A clear vision and strategy provide exactly the opposite. They define the frame. Humans continue to decide where the organization should go. AI helps illuminate where reality no longer coheres with that intent.
The assumptions that quietly disappeared. The pattern five teams are independently describing in different words. The recurring signal that fits no existing dashboard.
No executive can continuously hold an entire organization in working memory. A model can.
So my working hypothesis is this: With a clear vision and strategy, and a well-curated body of organizational knowledge, GenAI’s greatest contribution may not be replacing people at all. It may be supplying a layer organizations have never had — one that illuminates the dark corners of their awareness, helping leaders see strengths, risks, and opportunities that were always there but beyond the reach of any individual’s working memory.
I’m developing a framework around this idea, and I hold it as a hypothesis to test rather than a conclusion to defend.
If you’re exploring where GenAI can create lasting organizational value — or if your own experience points somewhere entirely different — I would genuinely enjoy comparing notes. Pick a time that works for you.
References
- MIT NANDA, The GenAI Divide: State of AI in Business 2025 — as reported by Fortune, Aug 18, 2025.
- Gartner press release, July 29, 2024; updated accounting per Gartner’s official statement: “By the end of last year, over 50% of GenAI projects were abandoned after the proof-of-concept stage.”
- S&P Global Market Intelligence / 451 Research, Voice of the Enterprise: AI & Machine Learning, Use Cases 2025 — as reported by CIO Dive, March 14, 2025.
- Gartner press release, June 25, 2025.