Internal · Working brief

Should we build an AI implementation service line?

A logical adjacency that doesn't deviate too far from our core, doesn't distract us, and leverages assets we already have.

Steve Patrizi. This brief is scored against that sentence.

The call

Yes, as a paid audit that converts into managed teams.

We'd lose a straight consulting fight. We have never built or bought services capability, the Big Four own the top of the market, and both model vendors now sell implementation themselves. Durable, affordable talent supply is the one position none of them can take, and we already own it. So the offer is a productized paid audit: 30 days, engineering orgs, existing customers first, converting into managed teams. We get in on the diagnosis and make our money on the talent layer.

What this is
Something to argue with. The outside research is done.
What's still open
One question can flip this recommendation. Section 08.
A no is fine
A deliberate no, backed by the homework, is a real outcome.
Sourced and verified Reported, or from an interested party Open, needs an answer from inside Revelo
02Why now

The top of this market is already taken.

A year ago the labs treated implementation as a side conversation. Now there are billions behind it, which is why this brief exists. That doesn't mean we should move fast, and section 08 says why.

$1.5B

Ode with Anthropic, an enterprise AI services firm founded by Anthropic with Blackstone, Hellman & Friedman and Goldman Sachs. Anthropic engineers sit inside the team.

$4B

OpenAI's Deployment Company, raised against a $10B valuation. Same play, larger. No investor overlap with Ode.

40,000+

Firms that have applied to Anthropic's Claude Partner Network. The badge differentiates nobody.

Anthropic is recruiting forty thousand partners and funding a $1.5B competitor to them at the same time.

What that does to the field

Both vendor firms place their engineers inside client operations, and both sell through their asset managers' portfolio companies. Hundreds of enterprises, already routed. Any independent has to assume the best logos go to the vendor's own firm first.

Where that leaves an independent

This narrows the field without closing it: mid-market rather than Fortune 50, and engagements that convert into something the vendor firms don't sell. That last clause is the whole recommendation, and section 05 tests it.

Sourced  Public reporting and vendor announcements confirm the Ode and Deployment Company figures, the embedded-engineer model, and the 40,000 applicant count. Sources at the end.

03What the research changed

Some of what we assumed at the kickoff didn't hold up.

This started from a handful of signals about where the market was going. Most of them hold up. The numbers kill one, and the partner network works in the opposite order from what we assumed.

Premise weakened

Turing grew on data work for the frontier labs

Turing tripled from roughly $120M to $300M, and most of that came from supplying data and evaluation labor to the frontier labs. Enterprise implementation was a much smaller piece. So what Turing shows is narrower than we assumed: a talent business can move up-market into higher-margin AI work profitably. Whether a staffing company wins at implementation is still an open question. Worth separating, because we already run the business Turing grew on.

Sequencing reversed

The partner network can't be the way in

Production deployments and published customer stories come out of delivered work and nowhere else. You don't join the network to get customers. You land and deliver customers, and the tier follows. Select sits within reach: ten certified people, two delivered customers, one public story. Global Premier belongs to Accenture and Deloitte.

Worry resolved

AI training gains volume from this

Tenex bundles upskilling as a workstream inside a transformation engagement. Training sits downstream of implementation, so the existing line gains volume from this. That removes one objection from the room.

Shape validated

Someone is already running this exact play from our starting point

Tenex, formerly 10x Labs, began in engineering services and software delivery, pivoted, and now sells a paid 30/60/90-day audit converting into outcome-based engineering pods. Same origin, same adjacency. Tenex is the closest thing to proof that the move is executable, and the clearest template for what phase one looks like.

Verify before it goes in a slide  The Turing revenue split comes from secondary analysis, and Tenex publishes its own engagement structure. Both are directionally strong. Neither is audited.

04The shapes this could take

Every shape this could take, including doing nothing.

These aren't mutually exclusive forever. B and C can grow out of A. Which one to start with is the decision, because each demands a different first investment.

Option
Uses what we have
Must build
What breaks it
AStart here

Entry point into staffing

A short paid discovery engagement opens the relationship, surfaces where AI fits, and converts into placements.

Revenue: placements · Fastest to first dollar

Uses what we have

Everything already running: recruiting, contracts, cross-border payroll, account management. The consulting layer stays thin by design.

Must build

A light discovery capability and one or two people who can run it. Smallest lift of any option.

What breaks it

If buyers want the work done rather than staffed, this reads as a sales tactic and burns trust. There's also a seam at the handoff: the executive who buys discovery isn't the VP Eng who buys engineers.

BThe losing option

Standalone retainer practice

A monthly retainer with an hours allocation, covering ongoing implementation. The HubSpot and Salesforce partner analogy.

Revenue: retainer · Slowest to first dollar

Uses what we have

The brand and the existing customer relationships. Little else.

Must build

An actual consultancy: methodology, consultants rather than engineers, change-management capability. We would be starting a second company.

What breaks it

We have no consultancy muscle, and we would fight the Big Four at the top, the vendor services firms in the middle, and thousands of certified boutiques at the bottom. On top of that, billing hours for AI-accelerated work is a trap: bill by the hour and you get paid less for finishing sooner.

CThe recommendation

Implementation into managed teams

The engagement lands, then we run a dedicated pod that owns the AI work on an ongoing basis. It sits between staffing and a services firm.

Revenue: managed pods · Medium to first dollar

Uses what we have

The talent pool, the cost structure and the client relationship. Closer to the core than a retainer practice.

Must build

Outcome ownership. Staffing places people into the client's team. Managed teams means we own delivery: team leads, project management, accountability for results rather than for hours filled.

What breaks it

When a placement underperforms you replace a person. When a managed team misses, we own the miss. That needs delivery leadership we don't have today.

DLead channel

Partner-network play

Join Anthropic's Claude Partner Network and use the co-marketing budget, brand association and co-events as a lead channel.

The honest read

This attaches to A or C rather than replacing them, which is worth stating plainly, because people mistake "become an Anthropic partner" for a strategy. The co-marketing money and enterprise co-events are real. The badge is table stakes.

And it can't come first

The tiers require delivered work. This channel switches on after something is already running. Getting the order wrong is the risk: mistaking a badge for a position.

ELive option

Do nothing

Focus on staffing and AI training, which work, and let this pass. The baseline every other option is scored against.

What would make this right

We can't bridge the consultancy gap fast enough. Or the market commoditizes before we establish a position. Or the distraction cost during a period that needs focus outweighs the upside.

If it dies

There's no shortage of high-impact work here if this dies. Section 08 lists what would make no the right answer. The cost of a no is missing the window while it's open.

A into C is the recommendation. Enter on the paid audit, convert into pods we run. B is where the money looks best on paper and the odds are worst.

05Can we win

The only position open to us is the talent layer underneath.

The market is good, and no one disputes it. The undecided question is whether we can take a defensible position in it with the assets we hold.

Ode sells forward-deployed engineers. The Deployment Company sells forward-deployed engineers. Accenture and Deloitte sell consultants. None of them own durable, affordable talent supply. We do.

Forty thousand firms can say "we do AI implementation." Ours is narrower: we find where AI pays in your business, and we are the most durable and affordable way to staff what comes after.

Assets

  • A vetted LatAm developer network, publicly stated at 400,000+, with cross-border compliance and payroll already solved
  • An existing frontier-lab AI business. Public reporting puts LLM training hires at 22% of FY2024 revenue confirm internally
  • Project-based delivery already in the catalogue. Managed teams is a shorter step than it looked at kickoff
  • Existing customers and trust, plus a US commercial footprint recently expanded by acquisition
  • A leadership team that already believes the market premise. This brief doesn't have to sell the diagnosis, only the response

Gaps

  • No consulting or change-management capability, and no history of acquiring any. Five acquisitions in thirty months, each one talent supply or US distribution. Closing this gap means changing direction
  • No strategist-led delivery. The audit phase isn't an engineering task, and everything downstream depends on it
  • Buyer mismatch. We sell to engineering leaders with hiring needs. Buyers for this sit a level or two higher reported
  • No delivery methodology. Turing has one, Tenex has one. Methodology turns services into something repeatable. Without it, each engagement runs on heroics
  • No bandwidth. No one here has room for this today

The alternative this brief has to name

Is implementation even the best next bet?

We already run the business that drove Turing's growth: frontier-lab data and evaluation work, publicly reported at 22% of FY2024 revenue, with relationships across the major AI companies. That is what Turing tripled on. Anyone on this team who knows those numbers would read a brief that skipped this as tunnel vision. My read: a narrow implementation play that feeds the talent business. A broad consulting practice would compete for cycles with a line that already works.

06What I'm seeing in the field

Someone just raised $12M to be the AI transformation company of Latin America.

Everything above came from desk research. What follows is what I've watched happen over the past year, including what landed this past week. I'm telling you where I think this goes, and I could be wrong about it.

The top end · August 24, 2026

Porsche handed its transformation to one vendor, and sold that vendor its own consultancy

Porsche signed a five-year, €1.25B agreement with Tata Consultancy Services to deploy AI across engineering, manufacturing, operations and customer experience. In the same deal, TCS is buying MHP, Porsche's own 4,500-person IT and management consulting arm, for €320M.

Look at the second half. Porsche sold its own consultancy to the vendor it just hired to do the work. At this end of the market, AI transformation is consolidating into vendors big enough to absorb the client's delivery organization.

The bottom end · opening now

Two-person operator shops selling judgment, with no bench behind them

cxo.dev launched this year: Claire Vo, a former CPO and CTO who built ChatPRD, with Zach Davis on engineering. Their line is Buying AI tools is not transformation. They sell a readiness assessment, then workflow redesign, then implementation support and measurement. No rate card. They describe themselves as a small team of product and engineering operators.

They're selling the judgment of people who have held the job. Shops like this are opening every month, and I expect a lot more of them.

At the top of this market a deal arrives with a 4,500-person acquisition attached. At the bottom, two operators sell judgment and have no bench at all.

The middle · August 25, 2026

Primero raised $12M to be the AI transformation company of Latin America

Primero launched publicly yesterday with a seed round reported as one of the largest in the region. It sells to Latin America's largest enterprises, its engineers work inside your systems and teams, and its leadership comes out of McKinsey, BCG, Goldman Sachs and Microsoft. On top of the services it sells a platform: an integration layer, a business ontology, agents, internal apps.

Read their positioning line carefully. Combining exceptional software with local engineering. That is our sentence, funded, and pointed at enterprises in our own region. They sell transformation to LatAm enterprises and we sell nearshore capacity to US companies, so the buyer is different today. They are hiring from the same pool and telling the same regional story.

$12M buys them a team. We have a network, which is a much bigger asset, and it is worth less every quarter nobody uses it.

The precedent

This is the agency wave again, and I ran a company through the last one

Between roughly 2003 and 2015, every company needed a website, then an app, and thousands of small shops opened to build them. I started Ideaware in that wave and ran it for fifteen years. The pattern had a shape: credible operator-led firms formed fast, the good ones got in on strategy and stayed for delivery, and the survivors owned something the next shop couldn't copy. Everyone else competed on price until the work stopped being worth doing.

AI transformation runs the same course, faster. The shops opening now are the 2005 web agencies. I didn't know how long the window stays open. Primero suggests that in our region it is already closing.

What it implies for us

We could be the bench for the shops that don't have one

Each of those operator-led shops generates implementation work it can't staff. Claire Vo can tell a company what to rebuild. She can't put eight engineers on it next month. We can, and it requires nothing we don't already run.

So phase 1 has two channels at no extra cost: sell the audit ourselves into existing customers, and be the bench behind the shops selling audits to everyone else. The first proves we can deliver. The second scales without us having to win the consulting fight we would lose. Both get harder once a funded regional player is hiring against us.

Where I think this goes  The Porsche, cxo.dev and Primero facts are sourced. The read on where this goes is mine, drawn from fifteen years of running an agency through the last cycle and a year of building with these tools daily. Worth pressure-testing before it goes to the leadership meeting.

07What it would take

The first move is small enough to approve in the room.

The first customer is us, and it takes no new headcount to reach the decision gate.

Phase 0 · No headcount

Implement on Revelo itself

Our own AI setup is the problem we would sell against: separate Claude instances per department, no shared context, no governance, likely overspending. Industry surveys find about a third of organizations have accurate visibility into their AI tooling and costs. We are a representative customer.

Run the engagement internally and we get a fixed internal problem, version one of a delivery methodology, real evidence about whether we can deliver this at all, and the first case study. It costs my time and no budget. And "we did this to ourselves before we sold it to you" is the most credible opening an implementation practice can have.

Phase 1 · Two customers

Productized, paid, 30 days

Fixed scope, fixed price, focused on the engineering org, sold into existing customers where relationship and trust already exist. The deliverable is a written adoption roadmap with quantified returns. Then convert at least one into an ongoing pod. Section 06 adds a second channel worth testing alongside it: one operator-led shop that has sold an engagement and has no bench to deliver it.

Repeatable, and priced on scope. If AI makes delivery faster and you bill by the hour, you get paid less for finishing sooner. We settle the pricing model before we sell anything.

Decision gate

Then we decide on evidence

Did the audits sell at the price we set? Did at least one convert into ongoing work? What was the margin? Answer those with real numbers, and we make the second decision, whether to build a practice, on real numbers.

The milestone is already defined for us

Phase 1 and the Anthropic Select tier are the same work

Select requiresPhase 1 produces
10 certified practitionersRegistration is free; exam fees still to be confirmed
2 production deploymentsThe two phase-1 customers
1 public customer storyWritten up from the better of the two

One effort covers both, and it settles the sequencing: deliver first, badge second.

Decide deliberately Select counts joint customers, meaning Claude deployed in production. The convergence holds if the phase-1 pods deploy Claude. Worth deciding now.

Resources

  • One owner who can run strategist-led discovery and write for an executive audience
  • A small engineering pod, drawn from existing bench rather than hired
  • Certification for roughly ten people
  • No new headcount to reach the decision gate
08Risks

The case against doing this.

The sharpest open question

Does delivery at the scale that matters require people on site?

Ode and the Deployment Company both put engineers inside client operations. If that becomes the expected shape of an engagement, our remote LatAm model inverts from its biggest advantage into a liability, in the segment holding the most money. Public sources don't settle it. This is the one finding that would change the answer in this brief, and the thing I most need help getting to.

RiskSeverityWhat it means
The model vendors are becoming competitorsHighOde ($1.5B) and OpenAI's Deployment Company ($4B) sell implementation directly, with vendor engineers embedded and PE-portfolio companies as pre-sold channels. The best enterprise logos route to the vendor firm first.
On-site delivery may be the expected shapeHighUnresolved, and the most important thing left to find out. See above.
Margins compress from the commodity endHighThe HubSpot and Salesforce ecosystems split in two: a few architecture-tier firms command premium fees while the rest watch margins shrink as vendors automate configuration. In AI this runs faster, because the models improve monthly.
The consultancy gap is structuralMed-highFive acquisitions, zero services capability. Closing it means hiring or acquiring a discipline we've never operated.
Buyer mismatchMed-highBuyers for this sit a level or two above our current motion. No one here sells at that altitude today.
A funded LatAm-native competitorMed-highPrimero raised $12M in August 2026 to sell AI transformation to Latin America's largest enterprises, with embedded engineers and ex-McKinsey and BCG leadership. A different buyer than ours today, the same talent pool and the same regional story.
Opportunity costMediumThe frontier-lab data business tripled Turing, and we already run it. Cycles spent here are cycles not spent there.
BandwidthMediumNo one has room today. A half-resourced services launch is worse than none.
Single-vendor dependenceLow-medBuilding a practice on one lab's partner program concentrates risk. OpenAI runs a parallel program and its own services arm.

What would make this a clear no

09What happens next

If the answer is go, here is what the first month looks like.

Nothing in the first month needs budget or new headcount. Here is the shape of it, and what the room would have to decide.

Week 1

A handful of short conversations

Where the frontier-lab data business stands today, what AI training sells and to whom, the customer base by vertical and size, and how we price project delivery now. I book them and I write them up. Point me at the right people and that is the whole lift.

The revenue one matters most. The only figure I have is 22% of FY2024 from a May 2025 article, and two acquisitions have closed since. This brief should not be quoting our own numbers from the press.

Weeks 1–2

Close the on-site question

A few calls with people running these engagements today. It is the one finding that would change the recommendation in section 08, and I would rather know before we commit than after. Steve offered his network for this at the kickoff, which is the fastest way to close it.

Weeks 2–4

Run the audit on ourselves

Map our own AI picture: which teams, which tools, what we spend, what governs it. This is the diagnosis we would sell, run on us first. It costs my time, produces version one of the methodology, and gives us the first case study before we have sold anything.

End of month

Named customers and a price

I come back with a shortlist, a fixed scope, a number, and what the internal audit found. That is the decision gate in section 07, and it arrives with evidence attached.

Before anything starts

What I'd need from the room

  • A yes on running the audit internally. Nothing else in the plan starts without it, and it is the cheapest thing on this page.
  • Names for those conversations. One introduction each, and I handle the scheduling and the write-up.
  • One customer relationship I can approach in month two. Existing, engineering-led, someone who already trusts us.

What you get at the end of it

A tested recommendation, our own AI spend under control, a delivery method written down, and a first case study. If the on-site question comes back wrong, you get a clean no in thirty days and we spent nothing but my time getting there.

Andrés Max · Revelo · August 26, 2026 Working brief · For discussion ahead of the leadership meeting