Being an AI-first company isn’t about buying AI licenses

A CIO buys 800 Claude licenses for the engineering team and announces the company is now AI-first. It isn’t. It has a Ferrari surrounded by sheep.

JOSÉ A. DORADO
CRO Orbitant

That was the image I used to open my talk at the Digital Enterprise Show 2026 in Málaga. And it captures what I wanted to say: being an AI-first company isn’t about handing out AI tools to your people — it’s about redesigning the system your business runs on. Raw power is useless when the process underneath is built to move at the pace of the flock.

In this article I share what I covered in that session: why most companies that call themselves “AI-first” aren’t, what layers actually need to change, and why human-designed processes remain the biggest bottleneck.

Buying licenses doesn’t make you an AI-first company

The CIO scene isn’t a caricature. It’s something I encounter regularly, and it almost always comes from the same combination of causes: FOMO, board pressure, the urge to shout “we’re AI” to the market, and — above all — a confusion between adoption metrics and impact metrics. How many active licenses, how many prompts per day, what percentage of the team “uses AI”… these are numbers that reassure the board but don’t move the business.

The problem isn’t the AI. The problem is that the org chart, the KPIs, the software delivery cycle, and the commercial cycle are all built to be operated by a human. When you plug a model on top of a process designed for a person, what you get is a local productivity improvement: the same task, slightly faster.

In the talk I framed it as two very different paths:

1. Tactical decisions (+100): layering AI on top of a human process to automate specific areas. This adds value, but it only creates a local peak — no real change. In the boardroom it gets sold as a “historic revolution”; in practice, it leaves the business where it was.
2. Strategic decisions (×10): the mindset shift that redesigns the system from the ground up. This is the only move that multiplies and changes the order of magnitude.

The difference between adding and multiplying isn’t how much AI you apply — it’s how you apply it. Adding means tactical decisions: I improve this report, speed up that response, automate an email. Multiplying requires a strategic decision, a fundamental change to how the company is structured. And that’s where the Ferrari stops being a joke: the tool is expensive and powerful, but the system around it is still moving at the pace of the flock.

This isn’t a fringe opinion. Companies that design from day one with agents in mind are already showing the way. Vercel launched Zero, a systems language built from the ground up to be consumed by agents — not just engineers. Stripe is working on infrastructure for agents to make payments programmatically, using single-use cards that never expose the customer’s real credentials. Neither is adding AI to what it already had. Both are rebuilding the foundation so that AI is a first-class citizen.

Being an AI-first company is an operating system, not a plug-in tool.

JOSÉ A. DORADO
CRO Orbitant

The four layers of an AI-first company

If being an AI-first company is an operating system, it helps to know which layers need rewriting. In the talk, I ordered them in four, from foundation to business, because real transformation means touching all of them — not just skimming the surface.

• Context: the company that answers back. Today, knowledge lives in people’s heads, in documents lost in Drive or SharePoint, in Slack threads and email chains. An AI-first company turns every action into an artifact the system can read: meetings recorded and summarised, reports and project status generated automatically, dashboards that connect every area. The uncomfortable question I put to the audience: is your company built for a human to read, or for an agent to query?
• Processes: humans who think, agents who execute. Every process must be able to capture information and improve itself. The what belongs to the human; the how belongs to the agent. As long as a process is still a sequence of steps designed for a person to execute one by one, AI can only act as an assistant — not as an autonomous system.
• Organisation: no human middleware. If you have a company that can be queried, is legible by AI, and is rich in artifacts, you don’t need people dedicated to forwarding information between teams. The focus shifts to thinking and building fast — prototypes instead of slide decks — and a new role emerges: something like HR for digital agents, managing their onboarding, their context, their integrations, and their evaluation.
• Product and business: Business Driven Development. Charging by the hour penalises speed. The delivery model must measure value, not time. And the engineer stops being a ticket executor: they capture context, model it, and orchestrate agents.

This layer ties back to something I repeated several times during the talk — the closer you are to the business and the customer, understanding metrics, context, and goals, the more the tool multiplies your impact.

Training isn’t just about improving skills — it’s about expanding context and judgment. That’s what determines whether talent adds or multiplies.

JOSÉ A. DORADO
CRO Orbitant

Training as an offensive strategy

If your product is your team, not training them now is the most expensive decision you’ll make this year. Because the transformation I’m describing isn’t something a procurement department buys — it’s something people build, day by day, getting things wrong and learning from it.

That’s why at Orbitant we treat training as an offensive lever, not an HR formality. Some of the formats we use:

• Non-technical team trained with Claude Cowork, so AI doesn’t stay inside engineering.
• Official Anthropic certification for the engineering team.
• Weekly internal and external sessions
• Twelve weeks of product-focused workshops.
• Asteroids: five-minute bursts to share one concrete learning.
• Super standups to give the whole team context.
• Fixed spaces to share, experiment, fail, and learn.

This connects directly to something we learned at a previous company (GuideSmiths): treat the team as the product. We discovered that technical excellence alone doesn’t maximise impact; without a business context, even the best teams hit a ceiling. Training isn’t just about improving skills — it’s about expanding context and judgment. It’s not an opportunity cost. It’s what determines whether talent adds or multiplies.

Start today, even though AI still fails

Today’s AI fails, loses context, and repeats mistakes. It’s a technology still maturing, but that’s no excuse not to start. That’s precisely why it makes sense to start now: waiting for AI to be perfect means giving up the time it takes to understand where it creates value, how to integrate it, and what capabilities need to be built around it.

The moat isn’t in the model — everyone has access to the same models, and they improve every month for everyone equally. The real competitive advantage lies in the months you’ve spent iterating with your team, accumulating context, and building the culture that lets you keep adapting when the tool shifts under your feet. Whoever starts today will have years of advantage over whoever waits for it to be “mature.”

I closed the talk with an invitation I’ll repeat here, because it’s the most practical one of all: before asking your team for a new AI plan, open an agent yourself and build something. If you don’t understand the tool first-hand, you can’t lead the transition. The transformation into an AI-first company doesn’t start with a purchase or a market announcement — it starts with the decision to redesign the system, and with the recognition that this decision belongs to you before anyone else.