Evidence — For and Against
The honest research on both sides of our bet — with sources. We lead with the case against on purpose.
Our whole plan rests on one claim:
The reliable money in enterprise AI is expert-led delivery — where a domain expert stays in control and approves every result.
A claim worth betting the business on should survive its strongest counter-arguments. Here is the research on both sides. We lead with the case against on purpose.
The evidence
FOR — most pilots fail; the winners embed
Most enterprise AI pilots show no measurable profit impact. The few that win build AI into work people already do — not standalone tools.
AGAINST — experts got slower
In a controlled study, experienced developers were slower with AI on their own code — while believing they were faster.
FOR — AI wins land on cost first
Cost-cutting automation pays back several times over. Revenue growth from AI lags far behind the hope.
AGAINST — the gains go to novices
The biggest measured speed gains went to beginners, not experts. Top performers gained little speed and lost a little quality.
FOR — AI is a mirror and a multiplier
AI amplifies what a team already is. Strong teams with solid foundations get real acceleration. Weak ones get faster chaos.
AGAINST — speed can cost stability
AI adoption links to less stable software delivery. Faster output can expose weak spots downstream.
FOR — foundations separate winners
Only a minority of infrastructure AI projects fully meet ROI expectations. Data quality and team support separate the winners from the stalled.
AGAINST — real AI products can still win
Packaged AI products get real Fortune 500 spend when they own a whole workflow. The weak play is the shallow wrapper — not every AI product.
A major enterprise AI-governance team puts it at roughly three-quarters of organizations adopting AI, with only a small fraction getting working ROI. The demand is real; most who chase it fail.
The honest take
The money is real, but it is narrow, fragile, and conditional. Most deployments fail. Experts can get slower when they over-trust the output. The biggest gains go to beginners. And speed can cost stability.
That evidence does not break our bet. It defines the conditions where the bet works — and those are the conditions VisiCore already meets:
This is why the pitch is not "AI makes everyone faster." It is "expert-in-the-loop, review-gated delivery" — the one version of the bet the evidence supports.
Sources
- MIT NANDA — The GenAI Divide: State of AI in Business — most pilots show no profit impact
- METR — AI's impact on experienced developer productivity — experts slower with AI
- METR — 2026 productivity experiment update — confirms the early result
- Brynjolfsson, Li & Raymond — Generative AI at Work — gains concentrated among novices
- DORA — State of AI-assisted Software Development — mirror-and-multiplier; stability findings
- a16z — Where Enterprises Are Actually Adopting AI — cost-first ROI; workflow-owning products win
- Gartner — AI projects in I&O stall ahead of ROI
- Internal enterprise briefing — the adoption-vs-ROI gap
Related: Objections & how we address them tracks every concern raised against the plan, with status.