The 97% That Isn't
Ask business leaders if they use AI and the answer is nearly universal. 97% of Indian SMBs now use AI tools in some form, and in the US, the Chamber of Commerce puts small-business AI use at 89%, up from 36% in 2023. Adoption, by this measure, is essentially done.
A different kind of survey tells a different story. The US Census Bureau's Business Trends and Outlook Survey — a government source with no product to sell — found that only 19.8% of all US businesses used AI in a business function, as of the period ending May 3, 2026. Among firms with four or fewer employees, that figure stayed under 20%, and didn't move at all over the prior six months, despite AI dominating every business conversation in that window.
"Using AI" and "AI changing how you work" are not the same claim. Nearly every business leader can answer yes to the first. Almost none can answer yes to the second. That gap is what this piece is actually about.
Five Postures, One Spectrum
These aren't five separate tribes of business leaders. They're five positions on a single continuum, and most leaders occupy more than one of them in the same year — sometimes the same quarter, depending on which part of the business they're looking at.
Why the Middle Is So Crowded
The cleanest explanation for why so many businesses sit stuck in the middle three postures comes from RAND's analysis of enterprise AI initiatives: more than 80% fail to deliver their intended value, roughly double the failure rate of an ordinary IT project.
That number tends to get read as a verdict on the technology. It isn't one. Behind the widely repeated "95% of AI pilots fail" headline sits a narrower, less dramatic fact: MIT's own research found the actual pilot-to-production success rate for task-specific AI tools was closer to 25% — five times higher than the sensational number everyone quotes.
What both studies agree on is where the failure actually happens: not inside the model, but in the layer of approvals, budget committees, and hand-offs between the person who wants something built and the team that has to build it. The bottleneck was never AI capability. It was the queue in front of it.
The Flip
Business leaders have started skipping that queue entirely. Instead of writing a spec and waiting on an engineering team's bandwidth, they're prototyping the thing themselves — handing it to a developer only once there's something concrete to harden, or, for a growing number of small businesses, shipping it without anyone else involved at all.
Gartner now tracks between 100 and 120 million people worldwide who regularly build business applications on no-code platforms, with no formal engineering background. In one 2025 industry survey, 82% of organizations said building custom software outside the IT department is now a core part of business strategy, not a workaround.
The cases are no longer hypothetical. One solo founder documenting her own build says she built her entire product frontend using AI coding tools at $225,000 in annual revenue, with no technical co-founder at any stage. A Singapore-based CEO with no engineering background told NUS Business School that six months after adopting AI tools, his company was running a fully custom CRM he built himself — no developers, no technical team.
The Graveyard Problem
None of this means the old risk disappeared. It moved. The same founder building Theanna warns against reading the shift as a purely technical unlock: the earlier generation of AI failures weren't failures of code either. They were failures to find a customer who actually wanted what got built, regardless of how skilled the team doing the building was.
AI removed the cost of building. It didn't remove the cost of being wrong about what to build. That's visible in the data from the "partly succeeding, partly failing" posture above — 65% of Indian startup founders surveyed by Elevation Capital say data quality is now a bigger bottleneck than AI capability itself. The leaders drifting between succeeding and failing aren't stuck because they can't build anymore. Building got cheap for everyone. They're stuck because judgment, clean data, and a clear question still take as long to earn as they ever did.
The Real Cost of Failing
Which is where the comparison that actually matters sits side by side. The average sunk cost of an abandoned enterprise AI initiative, by one 2026 estimate, is $7.2 million. The cost of a non-technical founder spending a weekend prototyping the same idea with today's tools is close to zero.
That gap is the real finding underneath every adoption number in this piece. The old way of trying and failing was expensive, slow, and required a committee's permission to attempt in the first place. The new way of trying and failing is cheap, fast, and requires nobody's permission at all — which means the businesses in the succeeding column aren't smarter than the ones stuck in the middle. They're just the ones who stopped waiting for someone else to build it for them.
The businesses in the succeeding column aren't smarter than the ones stuck in the middle. They're just the ones who stopped waiting for someone else to build it for them.
Sources
- LinkedIn Small Business Work Change India Report, 2026
- US Chamber of Commerce small business AI data, cited in AI Literacy in Small Business Statistics 2026
- US Census Bureau, Business Trends and Outlook Survey (BTOS), period ending May 3, 2026
- Elevation Capital, The State of AI Adoption in Indian Startups 2026 — CEO/Founder cohort
- BCG, India Insurtech Landscape and Trends: Reimagining Insurance with AI, Oct 2025
- The Agile Brand Guide, AI technology statistics compilation, 2026
- LegalZoom, AI and Entrepreneurship 2026
- RAND Corporation enterprise AI failure-rate analysis, cited via IEEE Computer Society
- Cooper, The AI Adoption Paradox, ISBM, Aug 2026
- Kissflow, No-Code Statistics 2026
- Index.dev, 50 No-Code and Low-Code Statistics for 2026
- Theanna — Building Theanna, founder's own account
- NUS Business School AI Showcase, Aug 2026