⚡ Key takeaways
- Consumer reach is huge: 44% of US adults had used ChatGPT by Feb 2026 (Pew, reported).
- Returns lag: 29% of companies report significant ROI from generative AI and 23% from agents (WRITER survey).
- Most agent pilots never reach production (about 88%, Forrester as cited); Gartner expects 40% of agentic projects cancelled by 2027.
- The blockers are integration (37%) and governance and security (34%), not model capability (Contentstack).
- Governments, regulated enterprises and the AI labs themselves value verifiable reliability most (our scoring).
🗺️ What the surveys say
Thesis 1: Reach is not the problem; production and proof are.medium confidence
| Finding (as reported) | Figure | Source |
|---|---|---|
| US adults who used ChatGPT / Gemini / Copilot / Claude | 44% / 24% / 17% / 6% | Pew, Feb 2026 ⚠️ |
| OpenAI weekly users | 900M (Feb 2026); 1B+ claimed later | OpenAI statements ⚠️ |
| Significant ROI: generative AI / agents | 29% / 23% | WRITER (2,400 respondents) ⚠️ |
| Top agent barriers | integration 37%, governance & security 34% | Contentstack (621 leaders) ⚠️ |
| Agent pilots that never reach production | about 88% | Forrester 2026, cited ⚠️ |
| Agentic projects cancelled by 2027 | 40% (forecast) | Gartner, cited ⚠️ |
| Mature agent governance | 21% of companies | Deloitte 2026, cited ⚠️ |
Samples, dates and definitions of "production" differ, and several surveys are vendor-run. The consistent signal: governance, integration and measurement block adoption more than capability.
So what: The gap between pilots and production is the market for anything that makes AI provable and governable.
📊 Who values verifiable trust
Thesis 2: Governments, regulated enterprises and the labs themselves.medium confidence
🔧 Reproduce this: customer_needs.py
customer_needs.py · our scores, informed by the surveys
"""customer_needs.py — which AI customers value verifiable trust most (our scoring).
Seven customer segments are scored 1-5 on five needs. Scores are OUR assumptions,
informed by the 2026 surveys cited in the article (governance and integration as top
barriers; quality as top blocker for agent teams). Maestro's fit = how much of each
segment's weighted need its assets (gate record, environments, on-prem) cover.
Standard library only.
"""
NEEDS = ["reliability", "compliance", "cost", "integration", "speed"]
MAESTRO = {"reliability": 0.9, "compliance": 0.9, "cost": 0.4, "integration": 0.7, "speed": 0.3} # how well Maestro serves each need
SEG = { # segment: scores per need
"Consumers": [2, 1, 5, 1, 5],
"Developers": [3, 1, 4, 3, 5],
"SMBs": [3, 2, 5, 4, 3],
"Enterprises (unregulated)": [4, 3, 3, 5, 3],
"Regulated enterprises": [5, 5, 2, 4, 2],
"Governments / sovereign": [5, 5, 2, 3, 1],
"AI labs (buying evals/envs)":[5, 4, 2, 2, 3],
}
res = []
for s, sc in SEG.items():
fit = sum(v * MAESTRO[n] for v, n in zip(sc, NEEDS)) / sum(sc); res.append((fit, s, sc))
print(f"{'segment':28} " + " ".join(f"{n[:6]:>6}" for n in NEEDS) + " fit")
for fit, s, sc in sorted(res, reverse=True): print(f"{s:28} " + " ".join(f"{v:6d}" for v in sc) + f" {fit:.2f}")
print("top three for a trust-and-verification product: " + ", ".join(s for _, s, _ in sorted(res, reverse=True)[:3]))$ python3 customer_needs.py
segment reliab compli cost integr speed fit
Governments / sovereign 5 5 2 3 1 0.76
Regulated enterprises 5 5 2 4 2 0.73
AI labs (buying evals/envs) 5 4 2 2 3 0.70
Enterprises (unregulated) 4 3 3 5 3 0.66
SMBs 3 2 5 4 3 0.60
Developers 3 1 4 3 5 0.55
Consumers 2 1 5 1 5 0.49
top three for a trust-and-verification product: Governments / sovereign, Regulated enterprises, AI labs (buying evals/envs)
Fit for a trust-and-verification product (0–1, our model)
| Governments / sovereign | |
| Regulated enterprises | |
| AI labs | |
| Enterprises | |
| SMBs | |
| Developers | |
| Consumers |
Real script output; scores are our assumptions.
So what: Speed sells to developers and consumers; proof sells to regulated buyers, governments and labs.
📄 Method and sources
- Survey figures as reported by the named organisations, mostly via secondary reports ⚠️.
- Segment scores are our assumptions informed by those surveys.
- Disclosure: Claude appears in the Pew figures; Claude, made by Anthropic, helps write this site.
Facts are sourced; figures marked ⚠️ come from secondary reports or company statements. Scores, weights and predictions are our judgement and are labelled. Disclosure: Claude, made by Anthropic, helps write this site. Anthropic is one of the companies analysed here, under the same rules as the others.
🧭 Foundations: the AI market, Nov 2026
Live data: AI Market Map & Investments · AI Investment Tracker.
Read
AI customers
You are here
Read