IT services in 2028: five decisions to avoid becoming a commodity

AI does not doom IT services firms. What it does is make a position that was already fragile far more dangerous: that of the interchangeable provider, paid mainly to mobilise resources on tasks that others can now accelerate, automate, or build directly into their platforms. The strategic question, then, isn't "how do we put more AI into delivery?" — it's "what will a client not be able to easily replace about us in 2028?"

The position argued here is this: an IT services firm won't protect itself by trying to become an AI lab. It will protect itself by becoming hard to substitute across five dimensions: business context, a control layer, reusable assets, a talent-development engine, and an orchestrator's position in the ecosystem.

This thesis is forward-looking. But it starts from signals that are already documented.

In brief

• •The French IT services market is not collapsing: Numeum and Xerfi put it at €34.6 billion in 2026, up 1%. That said, 49% of the IT services firms surveyed cite price pressure among their main headwinds, and 35% expected operating margin to decline in the first half.
• •AI has already entered delivery: in the 2025 Numeum-KPMG study, 72% of the IT services and technology consulting firms surveyed said they used it in their delivery processes. These are self-reported uses, not a measure of their effectiveness.
• •Productivity gains alone don't create an advantage: Numeum estimates they could rise from 15% in 2025 to 22.3% in 2027 among IT services firms, while stressing that they remain hard to convert into margin.
• •Model providers are themselves moving up the value chain. OpenAI has its Forward Deployed Engineers handle the scoping, design, build and production rollout of systems at strategic clients; Anthropic describes comparable engagements for its agent teams at large enterprises.
• •The data available shows neither the disappearance of the day rate, nor of juniors, nor the end of large generalist IT services firms. It points instead to differentiation through mere access to technical resources becoming less robust.

The real risk isn't AI. It's interchangeability.

It would be tempting to tell a simple story: agents write the code, clients will need fewer developers, day rates will fall, and IT services firms will have to find a new model.

The available data doesn't support such a mechanical conclusion.

In the first half of 2026, Numeum instead observes a return to slight growth in the French IT services market. The study does, however, highlight a more uncomfortable combination: cautious clients, pricing pressure, high technology and HR costs, and difficulty converting productivity gains into margin. 58% of the firms surveyed say they are repositioning around automation and AI.

In other words, AI acts less like a guillotine than like an accelerator of differentiation.

Two IT services firms can have access to the same models, the same coding assistants, and the same cloud. If one is essentially selling days while the other has mastered a business process, has proven assets, measures quality, owns the run, and knows how to arbitrate between several vendors, their competitive positions end up very different.

This shift is already documented in consulting: from production toward assets, proof, ongoing operations, governance and alliances, with an increasingly porous boundary between software, consulting and integration. Carrying this over to IT services firms produces the same tension: future economic rent probably won't sit in code production alone, but rather in context, integration, assets, operational accountability and trust.

It's from this tension that five decisions arise.

Decision 1 — Go vertical: choose the problems where you want to become hard to replace

What the facts show

Clients haven't stopped buying digital services. They're getting more demanding about the economic case for them. Numeum notes a growing role for price and for demonstrated ROI in purchasing decisions.

At the same time, a growing share of generic technological capability is being built directly into software, clouds and the models themselves — a reshaping of the value chain in favour of whoever controls the data, the platforms and the workflows.

What this weakens above all is the proposition:

"We can quickly supply you with skills on this technology."

Not because that skill becomes useless, but because it's increasingly easy to put out to competition.

Conviction

The first defence against commoditisation is vertical depth.

But "going vertical" doesn't mean tacking "Banking", "Industry" or "Healthcare" onto a practice's name. It means being able to take ownership of a specific value stream.

For example: cutting the processing time of a banking workflow; modernising a regulated application estate; improving the health of an industrial ERP; industrialising cyber-incident handling; making a data pipeline reliable in a constrained environment; running a customer service function under heavy regulatory scrutiny.

A vertical offering becomes genuinely differentiating when it combines four things: business expertise, technical architecture, relevant data or benchmarks, and the ability to measure an outcome.

The IT services firm is then no longer just selling the ability to build a solution. It knows what has to work, how to verify it, and what happens when it fails.

The limit

Going vertical isn't a universal recipe. Large IT services firms will keep drawing value from their international footprint, their staffing capacity, their certifications, and their standing with enterprise procurement. Excessive specialisation can also trap a firm in too narrow a market.

The question, then, isn't "generalist or specialist?". It's: on which subjects are we generalist out of necessity, and on which do we want to be irreplaceable?

Decision 2 — Build the control layer, not yet another model

One of the strategic risks easiest to misread is believing that the threat from Big Tech comes from the power of their models.

It also comes from their growing proximity to operations.

OpenAI now describes its Forward Deployed Engineers as responsible for deployments running from scoping through to a system in production, working directly with the client's technical and business teams. Anthropic, for its part, is hiring people able to run end-to-end agent deployments, define baselines and KPIs, measure ROI, and turn what's learned into reusable assets.

These are precisely the territories historically occupied by IT services firms and transformation consultancies.

This doesn't prove these players will replace them at scale. We have no data that would let us quantify such disintermediation.

But the signal deserves to be taken seriously.

Conviction

An IT services firm probably has no interest in entering the race for the best general-purpose model.

It does, however, have every interest in controlling what sits around the model: client context, connectors, permissions, agent orchestration, evaluation suites, security policies, observability, costs, execution logs, human-validation mechanisms, and the ability to swap out a model or a vendor.

The model can come from OpenAI, Anthropic, Mistral, Google or another provider. Operational control shouldn't disappear along with it.

This is where the IT services firm's future "control layer" lives: not a platform meant to rebuild everything, but a layer that guarantees the different AI components stay governable, measurable and replaceable.

The trap

Systematically building your own platform can become another form of running to stand still. A mid-sized IT services firm has no reason to rebuild functions its partners have already industrialised.

The right criterion isn't "build or buy".

It's: what do we need to own in order to stay in control of quality, of the client relationship, and of reversibility?

Decision 3 — Turn projects into assets, without turning assets into debt

When every engagement starts from a blank page, AI raises productivity, but not necessarily differentiation.

Any competitor equipped with the same tools can end up getting the same gain.

The picture changes when a project enriches a reusable body of assets: connectors, business-domain corpora, evaluation suites, specialised agents, architectures, rule libraries, migration accelerators, or operational dashboards.

Valuing and monetising these assets is one of the most well-documented strategic themes in the sector. Several cases of distributed capitalisation exist: at Stripe, according to an account cited by the Observatoire, teams publish agents for other teams to use; other cases highlight the growing importance of evaluations, internal platforms or reusable components. These remain individual cases, not proof that the model is universally superior.

Conviction

A 2028 IT services firm should be able to show what a 2027 engagement taught it, and what it will be able to reuse in 2029.

Otherwise, it stays condemned to reselling the same effort over and over.

But there's a fundamental difference between an asset and a folder full of proofs of concept.

An asset has an owner, a version, tests and evaluations, a known scope of use, a maintenance arrangement, explicit intellectual-property rights, a running cost, and a measured reuse rate.

A prompt left abandoned in a shared folder is not an asset. It's potentially debt.

The challenge is therefore twofold: capitalise far better, but be far more selective about what deserves to be maintained.

Decision 4 — Redraw the talent pyramid without breaking the senior-building machine

This is probably the most delicate decision, because a rational short-term optimisation can produce a long-term fragility.

Numeum notes that in the first half of 2026, 33% of the IT services firms surveyed had cut back on hiring recent graduates. But the same observatory is explicit that AI's direct effect on employment volumes remains limited at this stage, and that what AI is transforming first is roles, skills, and the organisation of work.

It would therefore be an overreach to write: "AI is eliminating juniors".

A more interesting conclusion emerges instead: across a panel of thirty cases selected for their organisational signals, with no claim to being representative, what stands out is a shift of work toward supervision and review, a hybridisation of roles, and some bottlenecks moving toward decision-making, specs and evaluations — with nothing in that body of cases demonstrating a general reduction in headcount.

Conviction

The strategic question isn't just how many juniors to hire.

It's: how does a junior become senior once AI absorbs some of the tasks they used to learn the job on?

An IT services firm can perfectly well shrink its base of repetitive production while keeping a learning pipeline intact. But that pipeline then has to be deliberately rebuilt.

That means, among other things: earlier exposure to incidents and client context, more testing and review work, learning to write evaluations and non-functional requirements, rotations alongside business experts and architects, mentoring time that's genuinely recognised, and bench time used to build skills and assets rather than simply wait for the next assignment.

The roles themselves are becoming broader: engineers turning into agent orchestrators, tech leads distributing work between humans and machines, and business-side profiles contributing more directly to building systems.

The wrong conclusion would be to infer that a 2028 organisation should be made up of seniors only.

Such an organisation might optimise its immediate delivery, but it would be outsourcing the production of its future experts to the job market.

Decision 5 — Choose your place in the ecosystem before others choose it for you

The value chain for digital services probably isn't going to simplify. It's going to get denser.

Around every client now sit model providers, hyperscalers, SaaS vendors, agent platforms, data players, cybersecurity specialists, IT services firms, consultancies, and the client's own internal teams.

At the June 2026 ESN & ICT Forum, Numeum was already describing the CIO–IT-services-firm–vendor–cloud-provider relationship as moving toward more alignment and partnership. The same forum noted that time-and-materials at a day rate was increasingly being challenged by capacity, outcome or gain-share arrangements — without going as far as announcing its disappearance.

A similar intuition emerges more broadly: mastering the ecosystem and its alliances is becoming a strategic component of positioning.

Conviction

Total independence is probably no longer a credible option. Neither is total dependence.

The IT services firm has to choose alliances that accelerate its value proposition while preserving a few essential capabilities: advising on several technologies when the client's interest calls for it, controlling the interfaces between the building blocks, measuring dependency on each vendor, organising reversibility, keeping its own knowledge of the client's business and information system, and being able to own the run when several vendors are pointing fingers at each other.

This is potentially one of the most defensible positions an IT services firm can hold: the third party able to orchestrate a system that no single other player fully controls.

The opportunity is real. But it comes with a requirement: no longer confusing partnership with simply reselling licences.

Challenge: five good decisions that guarantee nothing

This thesis has to be challenged.

First counter-example: size remains an advantage. Large generalist players aren't doomed for lacking specialisation. Their investment capacity, their vendor approvals, their geographic footprint, their talent base and their ability to take on large commitments remain powerful advantages.

Second counter-example: not every organisation is reorganising itself around AI. Companies like Lucca or Pennylane have integrated AI without overhauling their whole structure. The patterns identified in this body of cases remain lenses for reading a situation, not a universal target operating model.

Third counter-example: more assets can mean more cost. Every proprietary accelerator creates maintenance, documentation, security work, and sometimes lock-in. An IT services firm can perfectly well improve its competitiveness by leaning heavily on standard components rather than systematically productising its know-how.

Fourth counter-example: the day rate will keep rational uses. Scarce expertise, crisis situations, high uncertainty, transformations whose scope keeps shifting: several contexts remain poorly suited to a full outcome-based commitment.

Fifth counter-example: the client itself can bring things in-house. A good asset or platform strategy therefore doesn't guarantee the IT services firm will stay at the centre. One of the criteria for value will precisely be the ability to transfer skills without making every future relationship pointless.

These limits don't overturn the thesis. They sharpen it: commoditisation isn't fought by applying a single organisational model. It's fought by accumulating concrete reasons why the client chooses you for something other than a comparison of price and CVs.

Executive self-diagnostic: are we becoming interchangeable?

Score 0 for "no", 1 for "partially" and 2 for "yes".

• •Can we name three to five business or operational problems where our expertise clearly goes beyond mastery of a technology?
• •Do our main offerings have a baseline and metrics that let us demonstrate their value in a way other than the number of days supplied?
• •Can we switch AI model or vendor without rebuilding most of our delivery chain?
• •Do we have our own evaluations, control rules, connectors or context building blocks in our strategic domains?
• •Is an asset built on one engagement reused elsewhere, with an identified owner and maintenance budget?
• •Do we know the margin on our engagements after model, platform and licence costs, plus asset supervision and maintenance?
• •Does our junior hiring policy reflect thinking about building future seniors, rather than short-term headcount optimisation alone?
• •Do our managers know how to organise review, exceptions and decision rights between humans and agents?
• •Do our technology alliances increase our differentiation, or could they one day let our partner bypass us?
• •If a client asked us tomorrow "why you rather than the platform provider itself?", would we have a verifiable answer?

0 to 7 points, high exposure: your differentiation is probably still very tied to the people staffed and the technologies mastered.

8 to 14 points, transition under way: several building blocks exist, but they don't yet necessarily form a coherent economic system.

15 to 20 points, positioning under construction: the question becomes less "should we transform the model?" than "which building blocks now deserve to be industrialised and defended?"

The conviction

The IT services firm of 2028 will probably be neither a staffing company simply equipped with copilots, nor a software vendor in disguise, nor a lab building its own models.

The most robust position is probably somewhere in between.

It will sell the ability to achieve and sustain a digital outcome in a complex environment.

To do that, it will have to combine five advantages.

• •Business depth: understanding the problem well enough not to be selling technology alone.
• •A control layer: controlling context, evaluations, permissions, observability and reversibility, even when the models are third-party.
• •Reusable assets: turning accumulated experience into acceleration that's genuinely maintained.
• •A new talent pyramid: using automation without interrupting the production of tomorrow's experts.
• •An ecosystem strategy: drawing on dominant vendors without simultaneously ceding them the client relationship, the workflow and operational accountability.

AI doesn't necessarily strip the IT services firm of its value.

It forces the IT services firm to explain, much more precisely, where that value actually lies.