The value is real. The spend is real. And the gap between the companies getting one in exchange for the other and the companies getting neither has never been wider. That single line frames Hatchworks AI's State of AI 2026: Mid-Year Reality Check — and the numbers behind it are stark. The top 1% of firms spend $7,450 per employee per month on AI; the median firm spends $11. Enterprise AI investment hit $37 billion in 2025, tripling in a single year, and the boardroom question has sharpened from "does AI work?" to "show me the ROI."
For this special episode of Talking AI, host Matt Paige did something different: instead of publishing the mid-year report only in written form, the Hatchworks AI team used ElevenLabs to turn the full report into an audio experience. The voice is AI-generated, but the research, analysis, and point of view come directly from co-authors Brandon Powell, Matt Paige, and Omar Shanti. Here are the threads that anchor it.
If Your Organization Concluded the Technology Isn't Ready, That Conclusion Has Expired
January's "models are plateauing" consensus did not survive contact with the first half of 2026. The newest model generation holds long-horizon, multi-step work roughly three times better than models from six months ago, and Stripe used Fable to run a 50-million-line code migration — a project measured in engineer months — in a single day. The report's argument: the capability jump is the model multiplied by everything now wrapped around it, from agent harnesses to persistent memory to skills. Entire categories of work that were not automatable twelve months ago now are.
The ROI Variance Between Companies Has Very Little to Do With the AI
Everyone is buying roughly the same models at the same per-token price, yet one company clears the ROI bar and another does not. The difference lives in three places that don't come in the box: data connection, workflow embedding, and adoption. The MIT finding that 95% of pilots fail to hit the P&L makes the point — and so does its tell, that purchased tools and specialized partners succeed about twice as often as internal builds. AI success has always been a people, process, and integration challenge.
A Government Proved It Can Switch Off a Frontier Model by Decree
The defining story of the half: Anthropic's Fable 5 went dark for 18 days under a US export control directive, and what brought it back was identity — proving who is on the other end of the prompt. The report reads this as the arrival of trust-tiered AI, where access is gated by who you are rather than what you pay, and it lists four implications for enterprise buyers, starting with a blunt one: access is now a risk surface, and your business continuity planning should include the line "our model got turned off."
Open Source Is the Enterprise Hedge
A two-track market has matured: frontier models where the capability ceiling is the point, and open models that now cover roughly 80% of proprietary use cases at 10–20x lower per-token pricing. The report's recommended pattern is the advisor model strategy — frontier models plan, review, and grade while cheaper open models execute the bulk tokens. Coinbase is the proof point: CEO Brian Armstrong laid out five tactics — better defaults, routing, caching, lean context, and visibility over suppression — that cut the company's AI spend roughly in half while token usage kept growing.
The Firms Automating the Most Are Hiring the Most
Ramp's Economics Lab, working with Revelio Labs, paired observed AI purchases across 21,559 US companies with firm-level employment records. Heavy AI adopters grow headcount roughly 10% in the two years after adoption, and entry-level headcount grows 12% — exactly where the displacement fears have been sharpest. The gains accrue almost entirely to high-intensity adopters, though: companies that dabble see no statistically significant change. Depth is what pays.
The Bottleneck Is Still Human, and the Market's Answer Is a Role
Every thread of the report pulls toward the last mile — connecting systems, embedding workflows, encoding skills, and getting people to change how they work. The market's answer is the forward deployed engineer: senior builders embedded inside the business with a mandate of production outcomes, not slide decks. The report closes with nine calls for the second half of 2026, from ROI dashboards becoming standard to agent identity deciding deals — and a promise to grade itself in January.
Read the full written report: State of AI 2026: Mid-Year Reality Check
