Your people are already working through Copilot, ChatGPT and Claude, whether you planned for it or not. In the next decade, AI agents will replace websites and apps as the primary way work gets done. Every interaction, inside and out, will flow through AI.
Two decades of digital investment went into Customer Experience: designing how people discover, understand and transact with you. The next frontier is Machine Experience (MX): the same discipline, applied to machines, and pointed at one specific shift, taking AI from reasoning to action. It is how AI discovers your knowledge, understands your rules, and acts through your tools. Get MX right and you are woven into every agent and workflow your people, partners and customers use. Get it wrong and you are invisible to them.
From reasoning to action
Here is the core shift, and it is the whole game.
Today, most businesses use AI for reasoning. It answers, summarises, drafts and searches. Useful, but the work still lands back on your people. It is a brilliant thinking partner that stops at the edge of your systems, an expensive search engine that leaves you to do the doing.
The value is in the second step: AI for action. Give it reach into your tools and it does the legwork across your systems, pulling from your CRM, your project data, your documents and your finance system, drafting the output, and handing back for the decision. It works alongside your people rather than in place of them, and you stay in control.
That move, from AI that reasons to AI that acts, is what Machine Experience is for. Everything that follows is how you make it happen.
A new type of user has arrived
We wrote earlier about MX as designing for a new type of user, and it is worth sitting with. For thirty years we designed digital experiences for people: clear, intuitive, empathetic. But AI agents now interpret, decide and act, often autonomously, and they have become users in their own right. They browse your services, weigh your options, and choose whether to transact, much as a customer always has.
The catch is that machines do not experience your business the way people do. Humans need clarity, empathy and good design. Machines need structure, context and consistency. Where a person reads a page, an agent needs your knowledge to be labelled and discoverable, your processes exposed in a machine-readable way, and your rules and permissions carried alongside the data so it can act safely. Designing for that is a discipline in its own right, and most organisations have never done it.
Why most enterprise AI disappoints
Here is the uncomfortable truth: the AI on every desk cannot reach your systems. A chatbot bolted onto one application is a search box with better manners. It can reason about the data in that one box and nothing else, then hand the work straight back to you. Limited context, no ability to act. Impressive in a demo, thin in real work. This is why so many pilots stall after the applause. The model is rarely the problem; the problem is that it is marooned, cut off from the systems where the real work lives, stuck at reasoning when the value is in action.
The missing piece is tools. An AI surface is only as capable as the tools it can reach. Knowledge lets it reason and answer; tools let it act: check eligibility, run the numbers, submit the form, create the case. And the real multiplier is tools together. All your tools in one place, on the surface people already use, so the AI can chain them into a single flow. It stays safe because each person's own access travels with them, so the AI only ever sees and does what that person is allowed to.
It's not the tools. It's what they do together, where your people already work, governed by who they are.
Two problems, one cause
There are really two versions of this problem, and they share a root. Inside your business, the AI on your staff's desks cannot reach your systems, so it stays a clever conversation partner instead of a capable colleague. Outside your business, your customers and partners are beginning to arrive through their own agents, and if those agents cannot discover and use your services, they will route around you to a competitor who has made themselves reachable.
Picture a logistics company that exposes parcel tracking, scheduling and route optimisation so that any AI agent can use them directly. It opens an entirely new channel and becomes a preferred integration partner in the AI-driven supply chains forming around it. A rival that keeps those services behind a human-only portal quietly disappears from the same ecosystem. Same services, opposite outcomes. The difference is whether the business was built to be used by machines. And the fix for both the inside and the outside problem is the same.
Bet on the pattern, not the platform
The AI products are churning. Today's leader is next year's runner-up. Wire your tools into one vendor's assistant and you rebuild when it changes, and you are locked in until you do.
There is now an open standard that solves this: MCP, the Model Context Protocol. Introduced by Anthropic in late 2024 and adopted rapidly across the ecosystem, including through OpenAI's tooling, MCP is fast becoming the default way tools, data and services connect to AI. Think of it as the API for AI: the plumbing for the intelligent enterprise.
What makes MCP more than another integration standard is that it carries context, not just connections. Alongside each tool and data source it describes purpose, permissions and boundaries, so an AI system knows not only what it can do but what it is allowed to do, and for whom. That is what lets you expose your business once and have any AI surface use it safely: Copilot today, Claude tomorrow, your own agents next, a partner's agent after that. You own the capability. The products underneath can churn all they like.
Machine Experience, the discipline in four moves
MX is a repeatable method, not a one-off project. In practice it comes down to four moves.
- Design for the machine. Start with intent and governance. Decide what your knowledge, rules and tools should let AI do, who is accountable for the outcomes, and where the guardrails sit. This is where you treat AI as a first-class consumer of your business and set the parameters for how it participates, before a line of it is built.
- Consolidate business context. Bring your knowledge, rules and tools into a single governed place, your business context layer, rather than rebuilding access for every app, channel and pilot. This is also where the unglamorous data work pays off: content that is structured, labelled and discoverable, and processes exposed in a form a machine can actually use.
- Expose through the open standard. Publish that layer via MCP so any AI surface can use it, with permissions and boundaries carried alongside every tool. Standardising here is what turns a pile of point integrations into a single, coherent operating interface for AI.
- Meet people where they work. Deliver it on the surfaces your people already use, with each person's own identity and access deciding what the AI can see and do. Real work, in the flow, governed by who is asking.
Then it is simply applied and repeated: prove it on one high-value use case, measure, and scale. Every tool you add makes every surface smarter, because they all draw on the same layer.
What it unlocks
Sort this out and the returns compound. Add a tool once and it appears on every surface at once. AI crosses from reasoning to action: it does the legwork and hands back for the decision, safely, inside your rules. You govern by design, because access is inherited rather than bolted on. And you sidestep lock-in, because swapping or adding an AI product no longer means rebuilding your capability. It is one investment that spreads across every AI initiative you will ever run.
This is not theory
We are already delivering this kind of work in demanding, high-consequence environments. At the Fair Work Commission we built AI-powered case management that categorises and routes incoming matters and helps agents resolve them faster. At the City of Melbourne, a Microsoft Copilot pilot saved staff around 20% of their productive week, with 88% wanting to keep it. At ASIC, a governed data platform now lets AI help classify and triage regulatory cases, surfacing issues such as illegal phoenix activity earlier. Different sectors, same pattern: connect the business to AI, safely, and let it do real work.
Machine-first, or invisible
A decade ago, mobile-first design separated the leaders from the laggards. Machine-first will do the same in the decade ahead. The businesses being wired into the AI era are being wired in now, and the gap between "our people use AI" and "our business is usable by AI" widens every quarter. Early movers become the default tools inside their customers', partners' and staff's agents, and they help set the standards everyone else follows. Late movers spend years as an island the AI routes around, invisible to the systems increasingly deciding where and how digital interactions happen.
Where to start
You do not need a moonshot, and you should not start with one. Machine Experience rewards a narrow, deep first step over a broad, shallow one.
Start with the MX Starter: a fixed-scope engagement that takes you from an executive workshop to a live, governed tool in front of your people in about six weeks, with a costed roadmap to scale. You prove the pattern on one high-value use case, build the foundation you will reuse, and give your leadership a real thing to point at rather than another slide about potential.
Our view
Just as the web connected people to information and APIs connected systems to data, MCP will connect AI to the enterprise. Machines are becoming users: evaluating, choosing and integrating services just as customers always have. Designing for them, deliberately and with governance, is no longer optional. The organisations that master Machine Experience now will not just run more efficiently. They will future-proof how they operate, integrate and compete in an AI-first world, and they will help write the standards everyone else inherits.