The AI Value Fight at the Top, and Where Your Organization Actually Captures It
The argument going public
This month, Palantir chief executive Alex Karp took the AI fight public. On CNBC he said "something has gone completely wrong" between the AI labs and their customers. He accused the labs of hyping their models and overcharging for tokens, the units companies pay for as they use them.
Palantir followed with a white paper, "Institutional Sovereignty in the Age of AI." It listed fifteen steps companies could take to protect themselves from the likes of OpenAI and Anthropic. Karp put it bluntly: "at every single enterprise I deal with, these people are livid, they're like, 'I am paying for tokens that create no value.'"
He is not alone. David Sacks warned that the labs dominate the model layer, then move into the work their customers used to own. Microsoft's Satya Nadella was calmer but pointed: if you only consume a foundation model, he is not sure how you keep enterprise value.
Underneath the noise sits a more useful issue. It is where an ordinary business captures the value AI creates.
Most organizations are not capturing value yet
Start with an uncomfortable fact. Most companies get little from AI, no matter who they buy it from. A widely cited 2025 report from MIT's NANDA initiative found that 95 percent of organizations saw zero return on their generative AI pilots, against 30 to 40 billion dollars in spending.
That number is directional. It rests on interviews with 52 organizations and a survey of 153 leaders. The pattern behind it holds up across other data.
The study’s author attributed the results to a learning gap: companies had yet to rebuild their workflows, structures, or culture around the tools. In the same reporting, buying tools and partnering with vendors worked about 67 percent of the time, while internal build-it-yourself projects worked a third as often.
Other surveys agree. PwC's 29th Global CEO Survey, reported through Fortune, found only 10 to 12 percent of companies saw revenue or cost gains from AI, while 56 percent saw nothing. Treat that one with some caution, since PwC sells AI advisory work.
Government data tells the same story. Census Bureau figures from May 2026 put AI use near 17 to 20 percent of businesses nationally. Firms over 250 employees run at 37 percent, while the smallest firms sit under 20.
The frontier fight is about owning models, your fight is about changing the work
The executives are fighting over something most organizations will never need to win. Karp, Nadella, and Sacks are arguing over who should own the models and the knowledge inside them. That is a real question for a Palantir or a Microsoft.
Most organizations sit somewhere else entirely. Their value is trapped in daily work that never gets done fast enough.
Nadella is right that using a model is not the same as capturing value from it. Where we differ is on how that value gets captured.
That value comes from rebuilding real workflows around off-the-shelf tools, then making the new habit stick. The labs-versus-enterprise fight will play out on its own timeline, in courtrooms, pricing pages, and IPO filings. The work of turning AI into saved hours is available now.
What capturing value actually looks like
This is where field evidence beats argument. The organizations we work with use the same off-the-shelf tools everyone else has, applied to their own work. The results below come from changing workflows, not from owning technology.
Take Findorff, a Wisconsin general contractor. We built workflows for meeting summaries, RFI responses, and submittal review on Microsoft Copilot, ChatGPT, and Claude. Every person in the cohort was using AI after the engagement, and one project manager cut meeting documentation time by 40 percent.
Lehigh Construction Group ran about twenty workflows for estimating, invoice coding, and safety routing. Most staff saved 11 to 20 percent of their time. Safety routing alone dropped 45 to 60 minutes a day.
The same pattern holds in very different organizations. A heavy civil contractor in the Northeast cut payroll exception reporting from an hour to five minutes, and safety plan creation from a full day to two and a half hours. Idaho AGC took event follow-up from two to four hours down to fifteen minutes, and grew member check-ins from about ten a week to more than fifty.
The numbers hold across the board. AGC of Missouri measured 28.3 percent average time savings. The Tucson Metro Chamber cut policy monitoring time by 33 percent in four weeks, and AGC of Utah cut manual data entry by up to 90 percent on one workflow.
These are client-reported results and the numbers vary by organization. The through-line does not. None of these teams own a model, and every one of them captures value from off-the-shelf AI.
Where to start
The lesson under the headlines is practical. The teams capturing value focused on the work in front of them.
They took real work they already did and rebuilt it around tools they already had. They started with a few high-friction workflows, applied off-the-shelf AI, and gave people enough support to make the new approach stick.
That opportunity is available now. Find the workflows where time is being lost, where the same task repeats, or where people are buried in information. Rebuild one, measure what changes, then expand from there.
If your team is ready to stop watching the AI debate and start capturing value from it, we can help you find the workflows worth starting with. Explore the AI use cases that fit your operation, and ship the first one this quarter.