Summary. Anthropic and OpenAI just chose their distribution channel for the AI productivity wave: McKinsey, BCG, Bain, Accenture, Goldman Sachs, Blackstone. Two joint ventures, sixteen billion dollars committed, targeting a few thousand large customers worldwide. That channel will not reach the 3.44 million firms in the German Mittelstand. The constraint is not models or compute. It is the number of senior engineers willing and able to embed inside customer companies and build the AI capability into their workflows. A new tier of boutique implementers is forming to fill the gap. The 51 per cent of mid-market firms already using or testing AI (KI-Mittelstandsindex 2026) will buy from this tier, not from the frontier-lab joint ventures.
Three structural facts and one implication. AI productivity is rising sharply in the US, modestly in the EU, barely in Germany. The frontier labs have picked an enterprise distribution channel that bypasses the Mittelstand. The bottleneck for everyone outside that channel is human, not technical. Therefore the consultancy industry forks. The middle compresses. A new boutique implementer tier is emerging globally, and the Mittelstand will buy from this tier or not at all.
The Productivity Divergence Is Real
The a16z chart this week, drawing on Morgan Stanley research, shows US output per employee swinging from negative 0.5 per cent to positive 5.5 per cent in high-AI industries over twenty-four months. Output keeps rising while employment growth collapses, and the gap shows up as productivity.
The same axes plotted for Germany show the opposite movement. German high-AI productivity decelerates from +3.3 per cent in 24Q2 to roughly +1.5 per cent by 25Q4. The all-industries reading clears zero only late and at +0.6 per cent. The Bitkom 2026 study explains the mechanism: 41 per cent of all German firms now use AI, but the SME segment trails the larger cohort, and 53 per cent of all respondents name "missing competence" as the single largest barrier to adoption. This is the lag Germany lives in. Not zero, but visibly behind the European median and several orders of magnitude behind the US.
Frontier Labs Just Picked Their Distribution Channel
In February 2026, OpenAI launched Frontier Alliances with BCG, McKinsey, Accenture and Capgemini: multi-year partnerships in which OpenAI's own engineers work alongside consultancy teams inside enterprise customers.
In May, both labs went further. Anthropic announced a joint venture with Blackstone, Hellman & Friedman, Goldman Sachs and the Singaporean sovereign wealth fund GIC, capitalised at roughly 1.5 billion dollars (Anthropic press release, 4 May 2026). The target named in the release: community banks, mid-sized manufacturers and regional health systems with fifty million to one billion dollars of revenue. A week later, OpenAI launched the Deployment Company at a 14 billion dollar post-money valuation (TechCrunch, 11 May 2026), backed by nineteen private equity and consulting firms, starting with approximately 150 senior engineers acquired on day one to embed with clients.
Both labs are running the Palantir playbook: send senior engineers into the customer's own environment, configure the AI platform to that customer's specific workflows, and build switching costs one account at a time. The channel has been chosen. It points at the top of the mid-market, the upper mid-cap, and the largest enterprises in the Anglo-American axis.
It does not point at the German Mittelstand.
The Bottleneck Is Engineers Inside Customers, Not Models
Once you write down what one of these embedded engineers actually does, the supply problem is obvious. OpenAI's new Deployment Company starts with 150 of them globally. McKinsey's QuantumBlack, BCG X, Deloitte's AI practice and Accenture's AI group add to the pool, but no public estimate puts the total above the low thousands of senior implementers worldwide.
Compare that to the German Mittelstand. The Institute for SME Research (IfM Bonn) counts 3.44 million German enterprises, 99.2 per cent of all firms with turnover. Nineteen million people work in them. If the new frontier-lab joint ventures collectively allocate 5,000 senior implementers to Germany over the next two years (a generous estimate given the global headcount), the ratio is one implementer for every 688 Mittelstand firms. Those 5,000 will gravitate to the top thousand companies by revenue. The remaining 3.43 million are not on a slide deck in San Francisco.
The demand is four to six orders of magnitude higher than that supply.
A New Tier Is Forming
While the frontier labs were building their consulting joint ventures, a parallel tier was forming in the open market. Small, technically deep, builder-first firms staffed by senior engineers who configure AI directly inside customer environments. Different from Big Consulting. Different from generic software vendors. Different from staff augmentation. The clearest examples right now are American.
Tenex.co sells "AI transformation and AI engineering squads", targeting enterprises with organisation-wide AI strategy. Alephic embeds engineers inside marketing organisations at Amazon, Meta, PayPal, Ford, Disney and EY. Its tagline reads "Builders, not consultants. We solve problems by shipping code, not PowerPoints." Distyl AI has scaled the same model into a billion-dollar firm with outcome-based pricing, working with Fortune 500 customers in healthcare, telecommunications, insurance, manufacturing and financial services. Patterns AI, founded by ex-Anthropic engineers, sits in the same tier. Every combines an AI-native newsletter with consulting services and its own shipped software products, positioning itself as "AI training, adoption and innovation, from makers, not management consultants".
None of these firms look like McKinsey. None of them are trying to. They sell deep technical fluency, fast delivery cycles, and the willingness to embed engineers next to the customer's own people. The pattern is recognisable across all of them.
This tier exists in Germany too, in earlier and smaller form, focused explicitly on Mittelstand customers and EU AI Act compliance. The combination of frontier-lab supply concentration plus the regulatory deadline of August 2026 (high-risk obligations under the EU AI Act go live) means the German tier will either expand fast or leave the field to international firms that arrive too late.
How to Recognise a Real Implementation Partner
The Bitkom 2026 study names the three largest barriers to AI adoption in German firms: missing competence (53 per cent), data protection (44 per cent), and integration with existing processes (39 per cent). A real implementation partner has to dissolve those three, on contact. Three markers worth checking on the first call.
Do they understand your business? Can the partner describe your sector, your operating model, and the specific workflow they propose to change, in the language your operators use, before the second meeting? Generic "AI strategy" decks are the failure mode. Partners who cannot map a use case to your P&L within an hour will not survive contact with a real production environment.
Will they train your team? Schooling your own people is the largest single lever in adoption success, according to Bitkom. A partner who delivers a working system but leaves no internal capability behind is a one-shot expense, not a transformation. Ask: how do you build internal capacity, what does the handover look like, and what does your customer's organisation look like six months after you leave?
Can they handle compliance? EU AI Act Article 5 prohibitions went live in February 2025. High-risk obligations follow in August 2026. DSGVO is permanent. A partner who cannot name on the first call the specific regulatory artefacts that apply to your use case (AI Act risk classification, standard processing agreements, sector-specific frameworks where they exist) will become an external dependency the moment a regulator asks a question.
Underneath the three markers is one universal test: do they ship working software in weeks, not presentation decks in quarters? "Builders, not consultants" is not a marketing line. It is the only delivery model that compresses the time between board approval and operational impact down to the months your competitors are also working in.
What Follows for Founders, Boards, and PE
Mittelstand CEO. Do not wait for an Anthropic or OpenAI partner programme to call. The Monday action: identify the implementation partner who passes the three markers above and can put a working pilot into production inside ninety days. The legal form (independent specialist, partner network, small boutique) does not matter. The markers do.
Senior AI implementer (solo or boutique). Ten-year runway, not eighteen-month trade. Specialise in two or three verticals you actually understand. Document delivery patterns publicly. Treat vibe coding and rigorous evaluation as the two halves of senior craft, not opposite camps. Build compliance fluency as an explicit service line, not a footnote.
Mid-market PE investor. Re-underwrite portfolio companies in classical body-shop consulting at sharply lower multiples. Re-underwrite vertical boutiques and AI-native implementation firms with longer holds and higher multiples. The middle compresses; the extremes do not.
Board of a mid-market industrial. The next-meeting question is not "should we adopt AI" but "what is our internal AI implementation capacity, where is it sourced, and how does it compound". The 43 per cent of Mittelstand firms without a written AI strategy (BVMW Mittelstand Index 2026) are the ones whose competitive position is silently eroding right now.
The models are distributable. The engineers are not. Whoever reaches the Mittelstand reaches Germany.
How I Help
I run a small implementation practice from Munich. Builder-first, claude code native, German and English, focused explicitly on the three Bitkom barriers above and the August 2026 AI Act deadline.
The work runs in three layers, each designed for a different buying moment.
Diagnostic and assessment formats (under €1k). For the CEO, board member or operator who wants a concrete starting point in days, not weeks. An AI Readiness Assessment, an AI Act Risk Classification check, a Claude Code adoption review for a single team. Designed to be the kind of thing you can decide on without a procurement meeting.
Focused engagements (between €1k and €5k). Compliance sprints, half-day strategy workshops with the leadership team, use-case validations against your own data. Outcome is a written deliverable plus a decision: do we go further, and if so, on what.
Custom pilot engagements (scoped per case). One prioritised use case from discovery to production inside ninety days. Internal handover at the end so your own team owns and operates what we built.
More low-barrier formats are launching in the coming weeks as the entry-level catalogue grows. The fastest way to start, today, is a free thirty-minute discovery call. No obligations, no cold-call follow-up. Book at drfloriansteiner.com.
Forward this to a CEO, board member or PE partner thinking through the same questions. Subscribe at drfloriansteiner.com for next week's Weekly Agentic Engineering Digest on AI, claude code and the new vibe coding implementation tier.