The pitch in 2023 was simple. Buy the model. Plug it in. Get value. Two and a half years later, that pitch reads differently in the labs' own materials. We will build it. We will install it. We will sit next to your operations director for six weeks while she figures out where it actually fits. Sold separately.
That shift didn't show up in a single headline because nobody markets a strategic concession the way they market a model release. It showed up in two press releases eleven days apart. On May 4, Anthropic put implementation in writing. On May 11, OpenAI confirmed it. The substance is identical. Both companies are spending the kind of capital that doesn't move unless leadership has run the math twice. Anthropic put up $1.5 billion. OpenAI put up $4 billion and acquired a services company in the same breath. Neither of them needed to. Selling models is one of the highest-margin businesses anyone has invented in the last decade. They're doing this anyway, which means the math on "implementation as a separate revenue line" is now bigger, in their internal forecasts, than the math on "more API calls."
For Jacksonville businesses still treating AI as a tool you buy and turn on, the gap between the people who know what just changed and the people who don't is now wider than it has ever been.
The word "implementation" is doing heavy lifting in both announcements, and it deserves to be unpacked. When OpenAI says it's launching a deployment company, it doesn't mean someone is coming out to install ChatGPT on your laptop. It means a small team is sitting with the operations director for six weeks, mapping every place a human currently picks up a piece of paper, writes a similar email for the fortieth time, fills a form, or rereads a transcript. Then a model gets fine-tuned on the firm's own data. Then it gets wired into the workflow software the team already uses. Then it gets tested in shadow mode for two weeks before it ever touches a customer. Then a feedback loop. A metric. An owner.
None of that is what most Jacksonville businesses currently call "trying AI." Most Jacksonville businesses are in the ChatGPT-tab-open-on-the-side-monitor phase, which is genuinely fine for drafting an email and useless for compounding a competitive advantage. The labs just told you, in the most expensive language they have, that the second part is the only part that ever mattered.
A Jacksonville accounting firm doesn't need to read the TechCrunch coverage to feel this. They already see it in their pipeline. Their newer competitors aren't beating them on price. They're beating them on turnaround. The clients still want the same things they always wanted: clean books, useful insight, a person who actually picks up the phone. The firms doing it well are running document review against a private model trained on their own audit history. That isn't a tool you buy off a shelf. That's a system, built once, refined for months, watched by someone who knows the firm and the math.
Initial backing for the OpenAI Deployment Company, launched May 11, 2026, alongside the acquisition of applied AI firm Tomoro.
The honest read is this. The DIY ceiling is real, and it just got lower.
A solo operator with a paid ChatGPT subscription and three hours on a Sunday can still do real work in 2026. Draft proposals. Summarize meetings. Handle FAQ inbound. Run a research assistant before a meeting. That hasn't changed. What's changed is the gap between that ceiling and what a properly implemented system can do for the same Jacksonville business. The labs aren't building consulting arms because they pity the operators stuck on Sunday afternoons. They're building consulting arms because the gap is where the money is.
For a Jacksonville home services company, that gap looks like a model trained on three years of customer call transcripts that prequalifies leads while the office manager is still pouring coffee. For a Jacksonville law office, it looks like a discovery review system that flags privilege issues a paralegal would catch on a slow week and miss on a busy one. For a Jacksonville commercial real estate firm, it looks like a lease-abstract pipeline that turns four-hour readings into ten-minute reviews. None of those are tools. All of them are systems. All of them require an implementer, and the implementer's calendar is the thing that compounds, not the model behind it.
The relevant question for any operator reading this isn't "should we use AI." That ship sailed in 2024. The question is whether your version of AI is the side-monitor version or the system version, and what the cost of being in the wrong column will look like in twelve months. The labs themselves just placed that bet, in cash, with their own forecasts. The bet is that the cost will be larger than most operators currently believe.
Because the math shifted. In 2024, the labs assumed enterprise customers would buy the model and figure out integration themselves, the way companies did with cloud computing twenty years ago. That assumption broke. Most companies bought the model, ran a pilot, and stalled before production. The McKinsey number circulating in AI circles right now is that around 90% of generative AI pilots never reach scale. From the lab's perspective, that isn't a customer problem. That's a revenue ceiling. If your users can't get a model into production, they don't grow their API spend, which means your top line plateaus. The fix is to sell the implementation alongside the model, and to do it with the kind of margin a Big Four firm has been quietly enjoying for forty years. Both Anthropic and OpenAI made that call inside the same eleven-day window. Expect Google to follow within the year.
No, and anyone selling you that line is fishing. The DIY zone is still real for a wide range of tasks. Drafting outbound emails, summarizing long documents, generating first-pass marketing copy, building internal SOP libraries, prepping for meetings with a chat-based research assistant. A Jacksonville business owner can do all of that solo in 2026, and many do. What the consulting move from OpenAI and Anthropic signals isn't that DIY is dead. It signals that the ceiling on DIY is lower than the ceiling on a properly implemented system, and that the labs themselves now treat the gap as a permanent revenue line. The practical implication is calibration. If your AI use cases live inside one tab, you're fine. The moment a use case touches multiple data sources, customer trust, or compliance, the rules change, and a tab won't get you there.
A tool is something you open, use for a task, and close. A hammer. A calculator. ChatGPT in a browser tab. It does one thing well, and the value lives in the task you're doing while you're using it. A system is what happens when you stop closing the tool. A model fine-tuned on your own customer call transcripts and wired into your scheduling software is a system. It runs without you opening it. It compounds, because every call it touches makes the next one better. The lab consulting arms are not selling tools. They're selling systems. The signal for any Jacksonville operator is to ask, in the next quarterly review, which current AI use cases would survive if you forgot to log in for a week. The ones that wouldn't are tools. The ones that would, or that compound while you're gone, are systems.
Three quick tests. First, ownership. If a single person on your team is the only one who knows how the AI piece works, you're DIY. If three people could pick it up without a meeting, you're implementing. Second, integration. If your AI workflow involves a human copying output from one window into another window, you're DIY. If the model writes directly into the system of record, you're implementing. Third, the audit test. If you can answer the question "what did the AI do for us last Tuesday" without opening a tool, you're implementing. If you can't, you're DIY, and the model is doing work nobody on your team is currently watching. None of those tests require a consultant to administer. They require five minutes and an honest answer, which is what most Jacksonville operators have been postponing since the GPT-5.5 release last fall.
The labs just told you implementation is the product. Most operators haven't read the memo yet.
We help Jacksonville businesses move from the side-monitor tab to a system that compounds. One workflow at a time, with the kind of close-in attention the national consulting arms only sell at enterprise prices.
Book a Consultation Explore our AI services →