Every AI agency sells "no hype" now. "No bullshit." "Measurable results, not experiments." "Production-ready, not prototyping." The phrase used to mean something. In May 2026 it's commodity language: every consultancy says it, every landing page repeats it, and saying it tells you exactly nothing about who can actually build something that survives the next vendor pricing change.
I run a small AI and webdesign agency on Mallorca. I write this knowing I've used the same anti-hype phrases on our own site. They worked for about eighteen months. They've stopped working now, and the reason is worth pulling apart, because the deeper question underneath them is the one most prospective customers should actually be asking. It isn't about hype. It's about lock-in.
The Anti-Hype Class of 2026
The global AI consulting market hit $14B in 2026 and is projected to reach $116B by 2035, a 26% compound growth rate. That sounds like a rising tide. What the headline doesn't say is that the market is splitting hard. Deloitte and Accenture and Cognizant own the enterprise top. A long tail of boutique specialists owns the niche bottom. The middle, the generalist mid-sized agency that did "websites and a bit of AI," is disappearing.
Inside that splitting market, almost every surviving agency converged on the same marketing language. I checked the homepages of eight AI agencies on Mallorca last week, both German-speaking and Spanish-speaking. Five used the word "results-driven." Four led with "without the hype." Three used the exact same stock phrase, "production-ready, not prototyping," within a single screen of fold. One opened with "without experiments, with measurable results," which is the same sentence I had on our own services page eight months ago, almost word for word.
When everyone says the same thing, the thing itself stops working. Contentful's 2026 marketing study put it cleanly: "AI compresses time, but it also compresses differentiation." If anti-hype is everywhere, it's no longer a position. It's wallpaper.
What "No Bullshit" Actually Promises
Strip away the language and the implicit promise behind anti-hype marketing is this: we won't sell you something that doesn't work in production. Fair. Real. The problem is the unspoken second half, which is what kind of "working in production" they mean.
For most plug-and-play AI vendors, "working in production" means: we'll integrate one use case in two to four weeks, demonstrate measurable lift, and the lift will hold for the contract period. That is genuinely useful, and for some use cases it's the rational choice. Customer support agents that resolve standard tickets at scale, lead qualification flows with no industry quirks, internal copilots over existing documentation. Studies show 40 to 70% handling cost reduction in those scenarios when the buy-side integration is clean.
But the contract period ends. The vendor pivots. The pricing model changes. The second use case arrives, and it doesn't fit the first vendor's framework. And now the agency that promised you "no hype" is back at your door with another statement of work, because the architecture they delivered was never meant to extend. It was meant to ship.
That's not hype. That's also not honest about the trade.
The Real Question Is Vendor Lock-In, Not Hype
The thing customers should actually be asking in 2026 has almost nothing to do with marketing tone. It has to do with which layers of the stack the agency hands over and which layers stay inside the vendor's wall.
A recent framework from Expert AI Prompts breaks AI vendor lock-in into five layers that accumulate independently: model, orchestration, data, governance evidence, and organizational knowledge. Most plug-and-play deployments quietly lock all five at the same time. The model is the vendor's. The orchestration framework is proprietary. The embeddings live in their vector store. The audit trails are inside their compliance console. And the team that learned how the system works only knows that vendor's tools.
There have already been visible 2026 cases of AI platforms collapsing and taking entire enterprise deployments with them, but the framework names a harder and more common failure mode: "a pricing change at year 2 that the organisation cannot respond to because switching cost has accumulated to an unacceptable level." Orchestration lock-in is now the fastest-growing category of AI dependency risk. Most of the agencies selling "no hype" are precisely the ones routing their customers into it.
Cognizant's own enterprise research, which is hardly an outside critic of the consulting market, concluded that "plug-and-play artificial intelligence products fail to meet most enterprise needs". Buyers ranked custom solutions and flexible engagement ahead of pricing and speed. IT services firms, the ones who actually build and maintain rather than write strategy decks, had a 23% trust advantage over management consultancies. The trust gap is structural, not stylistic.
What Anti-Plug-Play Looks Like in Practice
If anti-hype is the wallpaper, anti-plug-play is the structural choice underneath. It's a less catchy phrase. It's also a more honest one, because it tells you what the agency is actually betting on.
For us at StudioMeyer, anti-plug-play means three concrete pieces of infrastructure we run ourselves rather than rent from someone. None of them are exotic. All of them are deliberate.
The first is our own memory layer. Every off-the-shelf chatbot forgets between sessions. Our hosted memory server keeps decisions, context, and patterns retrievable across weeks and months, and the same memory works whether the underlying model is Claude, GPT, or something local. A customer who builds on that memory keeps it when they switch providers. They don't have to retrain a new system from zero.
The second is custom MCP servers per customer. MCP is, in Anthropic's framing, "USB-C for AI." It's the protocol layer that lets any model connect to any tool through a standardized interface, so the integration outlives the model. We build a server tailored to each customer's stack, their auth, their data residency rules. The customer's own MCP endpoint is theirs. If Anthropic doubles their API price tomorrow, the customer points the same MCP server at a different model and keeps going. Aisera's 2026 build-vs-buy guide lists MCP, A2A, and AGNTCY as the three open standards enterprises should anchor on for exactly this reason.
The third is our own AI visibility tooling. We track how each customer's brand surfaces in ChatGPT, Claude, Perplexity, and Gemini. The data lives in our database, not the LLM provider's analytics console. If a vendor closes their developer API, the methodology and the historical baseline are still ours.
None of this is faster to deploy than plug-and-play. It costs more in week one. The trade is depth of fit and the right to leave.
When Plug-Play Wins
I'd be wrong to suggest anti-plug-play is always the answer. There's a genuine counter-argument and customers should hear it from anyone serious about advising them.
ServicesGround's 2026 analysis puts the custom-build ROI horizon at twelve to twenty-four months. Plug-and-play returns in weeks. If a company has a single bounded use case, no industry-specific data sensitivities, and a board that needs to see a number this quarter, buying beats building. 47% of enterprises already run a hybrid model where they buy where they can and build only where the work is genuinely differentiating. That's the honest pattern, not a binary.
The mistake isn't choosing plug-and-play. It's choosing it without understanding what it costs to leave it later, and being sold the choice by an agency whose entire model depends on you staying.
Four Questions to Ask Any AI Agency in 2026
If you're choosing an AI partner this year, four questions will tell you more than any anti-hype slogan.
What does the system look like in two years, when use case one is running and use cases two through five are arriving? An agency without a real answer is selling you the first sprint and not the architecture.
Who owns the infrastructure when this is live? If the answer involves the agency's proprietary platform, your switching cost just started accumulating in the first commit.
If the underlying model or vendor changes pricing, breaks an API, or goes out of business, what happens to your deployment? Vendor insolvency is the visible case. Quieter cases happen every quarter when pricing terms shift and the customer has no leverage to leave.
Are the data, the logic, and the knowledge portable? Schema, embeddings, audit trails, and the team's working knowledge should be things you can lift out. If they aren't, you don't have a system. You have a subscription.
Most agencies will not enjoy answering these questions. The ones who do are the ones worth working with.
What This Means
The AI consulting market is moving past a phase where saying "no hype" was enough. Kate Jensen, Head of Americas at Anthropic, put the broader frame to TechCrunch in February: "It wasn't a failure of effort. It was a failure of approach." The hype cycle is closing. The agencies still here in 2028 will be the ones whose customers can answer "we own this" rather than "we're locked in."
If you want to compare what plug-and-play feels like versus what owning the infrastructure feels like for your own case, our team in Palma will give you an honest read. Same offer if the answer turns out to be that plug-and-play is exactly right for you and we're not the right partner. The point of writing this isn't to win the pitch. It's to retire the phrase that stopped meaning anything.
