Tuesday morning, May 5, the AI press is still spinning the Anthropic news as a partnership. It is not a partnership. It is a pivot. Anthropic, Blackstone, and Hellman & Friedman each committed $300 million to launch a venture with one explicit purpose: embed engineers inside mid-sized companies and redesign the workflows around AI agents. Goldman Sachs, Apollo, General Atlantic, GIC, and Sequoia are along for the ride. The valuation came in at $1.5 billion before the venture had served a single client.
Twenty-four hours earlier, OpenAI's "Development Company" closed a $4 billion round at a $10 billion valuation. Same playbook. Different brand.
The labs at the absolute frontier of model capability are, this quarter, pouring billions into the people who walk into a company and rewire how it operates. The two firms with the best foundation models on Earth have looked at the market and decided that the next dollar should not go into more parameters. It should go into engineers who sit at the conference table and turn an existing workflow into an agentic one. They are, quietly, becoming the consulting firms that AI was supposed to disrupt.
That is the tell. When the people who build the models start selling implementation, you can stop wondering whether the bottleneck is the model. The bottleneck was never the model. The bottleneck was always the gap between an off-the-shelf agent and a workflow it actually runs.
This is what happens at the end of a gold rush. The first wave sold maps. The second wave sold picks and shovels. The third wave is selling pre-built sluice boxes that you operate yourself; and the people who used to sell maps have walked down to the river and started running them. Jacksonville businesses are not in California. Most are still standing on the porch reading about the rush in the paper.
Anthropic's joint venture with Goldman Sachs and Blackstone, announced May 4, 2026, exists for one purpose: embed engineers inside mid-sized companies to deploy AI agents into real workflows.
Last week's post laid out the PwC finding that 75% of AI's economic gains in 2026 are flowing to just 20% of companies. This week's news is the same finding from the supply side. The labs see the gap. They are now staffing for it. Anything an SMB can read into the price of NVIDIA, they should read twice into the staffing decisions of Anthropic and OpenAI.
The SMB data tells the same story from the demand side. 57% of U.S. small businesses are now investing in AI, up from 36% in 2023. Among committed adopters, 91% report direct revenue lift, and the average employee at an AI-using SMB recovers 5.6 hours per week. Adoption skews unevenly: 60% of SMBs with 50 to 499 employees use AI in at least one process, while only one in five firms with under ten employees do. The smallest Jacksonville businesses are the most underrepresented in the winning column. They are also the easiest to move.
And there is a real deadline this week. OpenAI's Workspace Agents have been free since launch. On May 6, tomorrow, the free runway ends and credit-based pricing begins. Any Jacksonville business that has been "meaning to try" the agentic stack has roughly thirty hours of free experimentation left. After that, deployment becomes a budget decision instead of a curiosity decision; and budget decisions, in most companies, take a month longer than they should.
Picture the operators this actually applies to. A Jacksonville HVAC company running six trucks, one dispatcher answering eighty calls a day, forty of which are basic scheduling that a voice agent handles in ninety seconds. A solo CPA in Riverside reviewing five thousand-line K-1s at three hours each, where an extraction agent finishes in eight minutes. A marketing agency on the Southbank pricing fourteen-hour research-heavy proposals as billable work, when the research itself is now a one-hour task. None of these operators needs a $1.5 billion joint venture. They need one workflow committed to production this month. The infrastructure is already cheaper than the staff hour it replaces.
The temptation, reading that Anthropic just put $1.5 billion into implementation services, is to assume the answer is also enterprise-scale. It isn't. The $1.5 billion is for Fortune 500 wiring projects. The Jacksonville version of the same move is one workflow, one owner, one metric, shipped to production within thirty days.
I propose a three-step framework that has worked across every Jacksonville client we've helped move from "exploring" to "running."
One: pick the workflow that costs the most repetitive hours per week, not the most interesting one. Owners default to the workflow they hate the most. That is rarely the workflow with the largest ROI. Add up your team's hours by category for one week. The biggest line wins. Boring usually wins.
Two: set a single metric you can prove in thirty days. Calls handled. Hours saved. Cycle time. One number. If you cannot say what success looks like before you start, you will not recognize it when it arrives.
Three: run it in production. Not as a pilot. Not as a proof of concept. Production means real customers, real data, real consequences, and a human in the loop on the high-stakes calls. The 86 to 89% of enterprise pilots that never reach production are not blocked by the technology. They are blocked by the word "pilot." Skip the word.
Cost is the most-cited barrier in the SMB data, with 61% of small businesses saying they cannot afford AI. That number is mostly wrong. The actual cost of one production agent in 2026 is dramatically smaller than one staff hour, often by an order of magnitude. The cost barrier is real for the consultants. It is rarely real for the agents themselves. The other barrier worth noting is security: only 10% of UK SMBs give staff any AI security training. Jacksonville businesses will eventually need a one-page policy on what data can and cannot leave the company through an agent. They do not need it before they ship the first one. They need it before the third.
The model was never the bottleneck. Anthropic and OpenAI just confirmed it by writing checks. The question for every Jacksonville business this week is not which model to use. It is which workflow goes into production by Memorial Day, who owns it, and what number you measure on June 5 to prove it worked.
The short answer is that they followed the data. For two years, both labs have watched enterprise customers buy access to frontier models, run pilots, and then stall at production. Industry research from late April found that only 11 to 14% of enterprise AI agent pilots reach scale; the other 86 to 89% sit in proof-of-concept limbo. The bottleneck was almost never model capability. It was the work of redesigning a workflow, integrating with existing systems, training staff, and operationalizing governance. That work is consulting work. Anthropic's $1.5 billion joint venture with Goldman Sachs and Blackstone, announced on May 4, exists explicitly to do that work inside mid-sized companies. OpenAI's $4 billion "Development Company" venture is the same bet. The labs concluded that the next billion dollars of value comes from putting engineers in conference rooms, not from putting more parameters in models.
OpenAI's Workspace Agents have been free since launch, running on OpenAI's dime as the company gathered data and refined the product. On May 6, that ends. Credit-based pricing kicks in, which means every action an agent takes will deduct from a metered balance. The change does not break anything. It does shift the decision from "let me try this for free" to "let me budget for this." For most Jacksonville small businesses, the actual dollar cost of running one production agent will be small, often a few cents per task. The bigger consequence is psychological: free experimentation runway closes, and getting started becomes a budget conversation instead of a curiosity conversation. If you have been telling yourself you'll set up an agent "soon," soon is today. Anything you stand up before the cutover keeps running, and your team builds the deployment muscle before the meter starts.
Pick the workflow that costs your team the most repetitive hours per week, not the workflow you find most annoying or most interesting. Owners default to the high-pain workflows because they remember those vividly. The high-volume workflows hide in the daily grind. For one week, have everyone in your business log their time by category in fifteen-minute blocks. The category with the most hours is the candidate. Then check three filters: it should involve structured data or text, it should already have a clear measurable outcome, and the cost of an occasional small error should be low or recoverable. Document review, customer inquiry triage, dispatch confirmation, lead follow-up sequences, and quote generation all consistently meet those filters in Jacksonville businesses. Resist the urge to pick the "cool" workflow. Pick the boring one with the most hours. The compounding ROI is in the boring one every time.
A pilot is anything where a human starts the agent each time, watches it run, and decides afterward whether to use the output. Production is when the workflow runs without anyone starting it; the agent fires on a real trigger, real customers see the result, and a human reviews only the cases the agent flags. The functional difference is whether the workflow happens with or without your team's daily involvement. If a customer fills a contact form and an agent drafts the response, routes it to a queue, and a staff member spends ninety seconds approving or editing before sending, that is production. If the same workflow requires someone to copy the form data into ChatGPT every morning, that is still a pilot. The 86 to 89% of stalled enterprise initiatives are stalled at exactly this line. The path across is one decision, made out loud, and a calendar deadline that the owner refuses to move.
The model was never the bottleneck. Implementation was. Anthropic and OpenAI just bet $5.5 billion on it. Jacksonville businesses can make the same bet at a Jacksonville scale, this month.
We help Jacksonville businesses pick the right workflow, ship a production agent in 30 days, and measure the result; without a six-month consulting engagement.
Book a Consultation Explore our AI services →