All services Every engagement has a fixed, published price. Book a strategy call
Enterprise architecture, explained in plain English: what EA is, what it is not, how it differs from solution architecture, and why in 2026 it governs software that acts as well as software that runs.
Ask what is enterprise architecture and you will get two kinds of answer. The textbook kind describes frameworks, layers and models, and leaves you none the wiser about what anyone does on a Tuesday. The honest kind is shorter: enterprise architecture (EA) is the discipline of deciding where change happens in your technology estate, in what order, and what it must not break. It connects business strategy to the systems, data and platforms that have to deliver it, so that every project builds towards one coherent estate instead of fifty private ones. That enterprise architecture definition has not changed. What enterprise architecture in 2026 has to govern has changed completely, because for the first time the estate contains software that acts on its own.
The short version. Enterprise architecture is a decision discipline rather than a documentation discipline. It exists to answer three questions before money is spent: where should change happen, in what order, and what must it not break. In 2026 its hardest job is new: governing AI agents, systems that do not just run but act. Gartner expects 40 percent of enterprise applications to include task-specific AI agents by the end of 2026, up from under 5 percent in 2025, and expects more than 40 percent of agentic AI projects to be cancelled by the end of 2027 for cost and risk reasons. The difference between those two numbers is, in large part, architecture. If your EA function still produces diagrams nobody uses in decisions, this is the year that stops being survivable.
Every organisation of any size runs on hundreds of systems, integrations, data stores and platforms, accumulated over years of projects that each made sense at the time. Enterprise architecture is the function that looks at that whole estate at once and makes it answer to strategy. In practice it does three things. It builds a shared picture of what exists and what the business needs it to become, usually anchored on a capability map rather than a systems inventory. It sets the target and the sequence: which capabilities get investment, which systems are strategic, which are tolerated, which are dying, and in what order the moves happen. And it holds the line while change is delivered: reviewing significant decisions, keeping projects on the agreed patterns, and stopping the estate from quietly fragmenting again.
Notice what is absent from that description: no framework worship, no wall of diagrams, no committee that says no for a living. Documentation is a by-product of enterprise architecture, the way minutes are a by-product of a meeting. The product is decisions. An EA function should be judged the way you would judge any decision-making body: were the calls right, were they timely, and did anyone follow them.
Half the failed EA functions we assess were built on one of these misunderstandings, so it is worth being blunt about them.
| The misunderstanding | Why it fails |
|---|---|
| EA is TOGAF certification | Frameworks are grammar rather than strategy. Knowing TOGAF 10 does not tell you which of your two ERP estates to retire, any more than knowing grammar writes the novel. |
| EA is diagrams and repositories | A repository nobody consults before a decision is a museum. EA tooling earns its licence only when it changes what gets approved and built. |
| EA is a review board that says no | Gatekeeping without a target state is just friction. Teams route around it, and the estate fragments anyway, now with added resentment. |
| EA is a job title, so we are covered | One architect with no mandate, no decision rights and no route to the investment conversation is an expensive way to feel governed. |
| EA is an IT concern | The expensive architecture decisions are business decisions in disguise: which markets share a platform, which products share data, what a merger keeps. Treating them as IT plumbing is how they get made by accident. |
The terms blur constantly, and the blur costs money, because organisations hire one when they need another. The distinction is scope and time horizon.
| Discipline | Scope | Horizon | Core question |
|---|---|---|---|
| Enterprise architecture | The whole estate: systems, data, integration, platforms, across every business unit | Two to five years | Where should change happen, in what order, and what must it not break? |
| Solution architecture | One system, product or project | The life of the project | How do we build this well, inside the enterprise guardrails? |
| Technical architecture | One technology domain: cloud, network, data platform | One to three years | How should this layer be designed and standardised? |
| IT strategy | Direction and investment posture | Three to five years | What role should technology play in the business plan? |
The relationship is simple: IT strategy sets intent, enterprise architecture turns intent into an estate-level design and sequence, and solution architecture delivers each piece inside those guardrails. Remove the middle layer and you get what most organisations have: a strategy deck nobody can execute and fifty well-built solutions that do not add up.
For forty years, enterprise architecture governed things that run: applications, databases, integrations. Predictable things. In 2026 the estate contains a new category, software that acts. AI agents read data, make choices and execute across systems, and they are arriving faster than most governance can absorb: Gartner expects 40 percent of enterprise applications to include task-specific AI agents by the end of 2026, against under 5 percent a year earlier. The same firm expects more than 40 percent of agentic AI projects to be cancelled by the end of 2027, mostly for escalating cost, unclear value and inadequate risk controls, which is a list of architecture failures wearing an AI costume.
This changes the job description. An agent is not an application you deploy and monitor. It is closer to a junior employee you onboard, permission and supervise. Someone has to decide what agents may touch, which systems expose which actions to them, how two agents with conflicting objectives are arbitrated, and how an agent is shut off when it misbehaves. Those are estate-level design decisions, and no individual project team can make them, because every answer constrains every other team. McKinsey’s 2026 work on agentic architecture reaches the same conclusion from the other direction: whether an organisation adds agents incrementally or rebuilds around them, the deciding factor is a deliberate, estate-wide architectural choice rather than fifty local ones. Regulation is arriving on the same timetable, with the EU AI Act’s transparency obligations live from 2 August 2026 and its high-risk regime from 2 December 2027, which we cover in detail in our guide to the EU AI Act and ISO 42001 in your architecture.
The uncomfortable conclusion for anyone who defunded their architecture function in the last decade: the discipline you cut for producing diagrams is now the discipline that decides whether AI creates leverage or liability. Our deeper analysis of that shift is in The Future of Enterprise Architecture in the AI Era.
Strip away the theatre and good EA leaves evidence. Five pieces, all checkable in an afternoon.
A capability map that appears in money conversations. Not laminated on a wall: open on the table when investment is prioritised. McKinsey found digital leaders use capability maps in 80 percent of relevant decisions against 38 percent elsewhere, and that their architects are in the room with executives at more than twice the rate. Decision records that name trade-offs. A written trail of significant architecture decisions, each stating what was chosen, what was rejected and why. This is also, by no coincidence, exactly what regulators and auditors now ask to see. A gateway that is fast enough to be used. Significant changes, and every new AI adoption, pass through a review measured in days rather than weeks. Slow governance is self-defeating because teams route around it. A register of what exists. Applications, integrations, data flows and, since this year, agents: what they may access and who owns each one. A roadmap wired to funding. The sequence of estate moves lives inside the investment plan rather than beside it. If the roadmap can be ignored without anyone noticing, it already is being ignored.
Yes, the way scales matter to a musician: as training and shared vocabulary rather than as the performance. TOGAF 10, the Open Group’s tenth edition standard, remains the reference point in 2026, and its modular structure is a genuine improvement on its predecessors. But no framework will tell you which ERP to keep after a merger or whether your agent platform should be centralised. Take the vocabulary, take the method where it helps, and never confuse adopting a framework with doing architecture. We compare TOGAF, Zachman and the rest, and when to use which, in our guide to enterprise architecture frameworks. The same logic applies to EA tooling: a repository platform amplifies a working decision process and embalms a dead one, so fix the process before you buy the tool.
| Situation | What EA does there |
|---|---|
| Major transformation ahead (ERP, cloud, merger, carve-out) | Designs the target and sequence before contracts are signed. This is the cheapest moment to be right and the most expensive to be wrong. |
| AI adoption accelerating | Decides the platforms, guardrails and agent permissions once, so fifty teams do not improvise fifty answers. |
| IT spend rising with nothing to show | Finds the duplication and the systems kept alive by habit, and gives the board a defensible simplification plan. |
| Every project is an integration surprise | Replaces point-to-point improvisation with deliberate integration architecture. |
| Regulated, or about to be | Produces the evidence trail (decisions, data flows, AI register) that DORA, the EU AI Act and ISO audits now demand. |
Size is the wrong test. A 300-person firm mid-ERP-selection needs architecture judgement more than a 5,000-person firm in steady state. What smaller organisations do not need is a permanent EA department, which is why fractional models exist.
The classic failure mode is maturity-model thinking: two years of foundations, frameworks and tooling before the first useful decision. Invert it. Start with the next irreversible decision your organisation faces, and do the architecture for that: the capability view, the options, the trade-offs, the recommendation with its evidence. Then the next one. A function that earns credibility one good call at a time gets its mandate extended. A function that asks for two years of patience gets defunded, and on current evidence deserves to be. Our practical EA guide for CIOs sets out the first 90 days in detail, and our guide to EA operating models covers how to structure the function as it grows.
Intelance is a UK advisory firm that does enterprise architecture for a living, at fixed, published prices. The EA QuickScan, from £18,000 + VAT, is a senior-led assessment of your estate and architecture function with a sequenced plan at the end. The Virtual EA Department, from £6,500 per month, gives you a working architecture capability without the £180,000 hire. The full service range is on our enterprise architecture consulting page.
It is the discipline of deciding where change happens in an organisation’s technology estate, in what order, and what it must not break, so that individual projects add up to one coherent whole instead of accidental complexity.
They maintain the shared picture of the estate and the target it is moving towards, shape and review significant technology decisions before money is committed, and keep delivery teams on agreed patterns. The output that matters is decisions rather than diagrams.
Scope and horizon. Enterprise architecture governs the whole estate over a two-to-five-year horizon. Solution architecture designs one system or project inside those guardrails. Most organisations have plenty of the second and none of the first, which is why well-built systems still add up to a mess.
No. TOGAF 10 is a framework: shared vocabulary and method. Enterprise architecture is the decision-making practice itself. You can run excellent EA with a light touch of TOGAF and terrible EA with a room full of certified architects.
Architecture now governs software that acts rather than just software that runs. AI agents need estate-level decisions about permissions, platforms, arbitration and shut-off that no single project team can make. Gartner expects 40 percent of enterprise applications to include task-specific agents by the end of 2026, and expects over 40 percent of agentic projects to fail by 2027, largely on governance and cost grounds.
It needs the decisions rather than necessarily the department. A firm facing an ERP selection, a merger or serious AI adoption needs architecture judgement at that moment. Fractional models such as a virtual EA function provide it without a permanent hire.
Look for evidence in decisions: a capability map used in investment conversations, written decision records with trade-offs, a review gateway fast enough that teams use it willingly, a current register of systems and AI agents, and a roadmap wired into the funding plan.
A permanent function starts at roughly £180,000 per year for one senior hire before tooling. Fixed-price alternatives exist: our EA QuickScan assessment runs from £18,000 + VAT, and a fractional Virtual EA Department starts at £6,500 per month.
Written by Emmanuel Olatunji, Chief Enterprise Architecture Partner at Intelance. Based on two decades of architecture practice across UK enterprises, Gartner and McKinsey research published for 2026, and the TOGAF Standard, 10th Edition. Last reviewed 8 July 2026.
Emmanuel Olatunji
Chief Enterprise Architecture Partner
Emmanuel founded Intelance in 2017 and has led architecture, AI governance, cyber and deal engagements across the UK, Europe, the Middle East and Africa. Every Intelance engagement runs under his direction, and every guide he publishes comes from that work.
See his record on LinkedIn About IntelanceThirty minutes with the architect who wrote this guide. A clear recommendation and a fixed fee in writing, within one UK business day.
Talk to a senior adviser
Enterprise Architecture
Enterprise Architecture
Enterprise Architecture