Build, Buy, or Boutique: Who Should Run Your Agent Strategy?

When a mid-market company decides to get serious about AI agents, the partner question has four real answers: build the capability internally, hire a major consulting firm, take the strategy your software vendor offers, or bring in a boutique operator. I sit in the fourth category, so discount accordingly — but I am going to give you the honest version of all four, including the situations where the right answer is not me.

Option one: build it internally. This is the right answer more often than consultants admit — specifically when AI is becoming core to your product, when you already employ strong platform engineers, or when the work is so entangled with proprietary systems that no outsider can hold the context. The honest costs: the talent is scarce and expensive, the learning curve is twelve to eighteen months, and your new team will spend much of year one rediscovering lessons that are already published. Internal builds also struggle with the diagnosis phase — it is genuinely hard to assess your own organization's task structure from inside it, because everyone inherits the org chart's assumptions about where the work is. A reasonable hybrid: buy the diagnosis, build the execution.

Option two: the Big Four and the big strategy houses. What you are buying is real: scale, methodology, regulatory cover, and a brand your board already trusts — nobody gets fired for hiring them, which is precisely what they price. The honest costs for a mid-market buyer: engagements built for enterprise budgets and enterprise timelines — typically six to twelve months and well into six figures; delivery teams staffed with junior consultants while the partner who sold you appears quarterly; and research that is survey-based rather than operational. Their published insight comes from asking thousands of executives what they think is happening, not from running agents in production. One more structural fact worth knowing: the big firms are arming for the enterprise agentic market at extraordinary scale — PwC and Anthropic, for example, announced an expanded alliance in May 2026 that includes training and certifying 30,000 US professionals on Claude. That tells you exactly where their attention and economics point: upmarket. A 1,500-person manufacturer is not the client those practices are being built for.

Option three: take the vendor's strategy. Every agent platform vendor will happily assess your organization, often free. Understand what you are getting: an assessment whose conclusion is known before it starts. The vendor's diagnostic exists to sell the vendor's platform, and it will find — every time — that your highest-value opportunities happen to match their product's strengths. That does not make it worthless; vendor teams know their own tooling deeply, and a free assessment can be useful input. But it is input, not strategy. Nobody whose paycheck depends on the answer should be the one framing your question. The tell is org-neutrality: ask the assessor what they would recommend if you bought nothing from them. Watch what happens.

Option four: the boutique operator — a senior practitioner who does the work personally, runs agents in their own business, publishes a price, and has no platform to sell. The honest advantages: you get the actual expert rather than the expert's slide template; the economics work at mid-market scale; and operator evidence — what the tools actually did in production — beats survey evidence for the questions that matter to you. The honest disadvantages, because every option gets the same treatment: less scale, so a boutique cannot parachute forty people into a global rollout; no brand cover, so you carry the internal sale; and wildly variable quality across the category, so diligence is on you. The boutique lane has real operators and it has people with a deck and a Calendly link, and they look identical on LinkedIn.

So how do you cut through? The same questions work on all four options, me included. Who, by name, does the work — and do I get their time or their team's? Does the assessor run agents in their own operation, or only write about other people's? Is the price published, or does it emerge from a scoping process calibrated to my budget? Will they name the parts of my business where agents will not work? And do they quote market statistics with the sample frames attached — because a partner who tells you "88% of businesses use AI" is either careless or selling, and that statistic is actually 88% of respondents to a self-selected McKinsey enterprise survey, while representative US Census data puts AI use at roughly one in five US businesses. Statistical carelessness upstream becomes capacity-math carelessness downstream, on your budget.

Whatever you choose, weigh it the way I weigh any significant investment in my own companies: is there a real market reason to act, does the specific plan make sense, and what is the true investment of time, effort, and resources to realize the upside? Partner selection is just that filter applied to people. The Big Four clear the first test trivially and get expensive on the third. Vendors are cheap on the third and compromised on the second. Internal builds win the second and bleed on the third. Boutiques have to prove all three — make them.

Context that should shape the decision: this market punishes unclear value fast. Gartner forecast in June 2025 that over 40% of agentic AI projects will be canceled by the end of 2027 — an analyst forecast, not a measurement, but it names the killers: escalating costs, unclear business value, inadequate risk controls. Score any partner on exactly those three. Will they cap and stage your costs? Will they give you a defensible value number before deployment, with conservative math you can challenge? Will they leave you with working governance — decision boundaries, monitoring, audit trails? On that last one the bar is low and the stakes are not: Deloitte's 2026 survey of 3,235 enterprise leaders found only 21% reporting a mature governance model for agentic AI. Demand the governance deliverable in writing, from anyone.

My rubric, by situation. AI is your product, or you have deep platform engineering: build, and consider buying only the diagnosis. Regulated enterprise, 10,000+ employees, board needs a name brand: the big firms exist for you, and their economics finally make sense at your scale. Committed to a single agent platform with high confidence: take the vendor assessment as free input, then get an org-neutral check on it before you re-architect work around it. A 500-to-2,500-employee organization that needs to know where its real opportunities are, sized in dollars, before committing serious budget: that is the boutique-diagnosis profile — and yes, that is the lane I built SDI Clarity's assessment for.

What we are, stated plainly so you can apply the same scrutiny to me: a five-week agentic workforce assessment starting at $25,000, tiered by organization size and published right on the site; the work done personally by a practitioner who runs agents in his own companies daily; no software to sell you, no implementation army waiting behind the diagnosis, and a deliverable designed to be argued with — conservative math, stated assumptions, board-ready. What we are not: a platform vendor, a body shop, or the right answer for a 40,000-person global rollout.

I publish the price and the method because I am open to being challenged — always — as long as it is a fact-based conversation, and a published price is where fact-based conversations start. Whichever lane you pick, pick it with called shots and measured results, and switch lanes the moment the evidence says you chose wrong. The partner is a means. The capacity is the point.