Three months since.

We've been on the AI journey at Travelopia for two years now. Back in May, we started writing publicly about adoption—the behaviour change problem, the gap between tasks and transformation, the mess that is "both in the room." Three months on from that last post, we're seeing which of those ideas have held up and which needed refinement. This is what two years in actually looks like.

The behaviour change thing? Still true.

We said that AI adoption fails when it's treated as a technology rollout instead of a behaviour change. That hasn't shifted an inch. If anything, it's become more obvious.

What's changed is what that actually means in practice. We thought behaviour change was primarily about leadership visibility and personal example—and that's still true, leadership sets the ceiling. But we've learned it's more granular than that. It's not just "leaders need to use AI." It's about the specific routines and rituals that embed AI into how a team works. The daily standups that reference what AI helped with and what didn't. The code reviews that normalize AI-assisted code. The retrospectives that ask "where did AI miss for us this week?"

Behaviour doesn't shift through inspiration. It shifts through repetition and feedback loops.

Tasks to transformation—we were right, but it takes longer.

We wrote about how most organisations are getting better at tasks but not at transformation. The trap is that early wins feel like success, so people stop there. "Think small" becomes the new normal.

That's still accurate. But there's a pattern underneath it we've noticed: the shift from tasks to flow isn't a flip. It's a gradual expansion. You start by automating a single task. Then you notice the task that feeds into it could be better connected. Then you see the decision that depends on both. Over months, what started as one small improvement becomes a different way of working—but only because you kept pointing at the friction and asking why.

The organizations making this shift aren't the ones with a grand plan. They're the ones that get curious and stay curious.

Agents and automation: know the difference

There's more hype around AI agents now than there was three months ago. Agents that work independently, make decisions, take action. It sounds revolutionary. But we've learned something important: more autonomous doesn't mean better.

Automation works brilliantly for well-defined tasks with fixed rules. The steps never change. The outcome is predictable. Machines excel at this. But most valuable work isn't like that. It requires judgment. It involves tradeoffs. Someone needs to decide what matters.

AI agents are getting better at handling complexity. But they still need guardrails. They need humans checking their work. They need feedback loops. They need to know when to escalate.

The organizations we're watching that are succeeding with agents are being surgical about where they deploy them. Not everywhere. Not because the technology exists. Only where the agent is solving a real problem better than the alternative.

That's actually harder than it sounds. It means resisting the hype and thinking clearly about what you're actually trying to solve.

The triangle: all three corners have to move

We wrote about "both in the room"—technology and people together. We've realized there's actually a third corner that matters just as much, and most initiatives miss it.

Think of it as a triangle:

                    BUSINESS
                   /        \
            Clear ROI    Real Value
           Accountable   Measurable
                /            \
               /              \
            TECH ----------- PEOPLE
         Systems Ready    Understand Change
         Data Sound       Confident & Trained
              \              /
               \            /
              All Three Moving Together

Technology is one corner. You need solid infrastructure, reliable systems, good integration. That's necessary but not sufficient.

People is the second corner. Teams need to understand what's happening. They need to be trained, supported, confident. They need to buy in to why this matters. And they need to see how AI actually changes their work—whether that's freeing time, making things clearer, or shifting what they focus on.

Business is the third corner. There need to be real business drivers. Clear benefits. ROI that can be articulated. Someone accountable for the outcomes. And leadership that's aligned on what success looks like.

Most initiatives stumble because they're optimizing for one corner and ignoring the other two. You can have brilliant technology and zero business impact if people don't use it. You can have enthusiastic people and great change management but fail if the technology doesn't hold up. You can have clear business goals and good intentions but still stall if the people side isn't ready.

All three corners moving together, not taking turns.

The organizations that are actually winning are treating this as a genuine triangle. Tech leaders working with business leaders. Business leaders talking to the people who'll actually use the systems. People having a voice in how technology gets deployed. It's messier than separate workstreams. But it's the only version that consistently works.

Cost-to-value: where real impact is starting to show

Three months ago, organizations were still in the "invest in everything" phase. Try all the tools. Deploy everywhere. Optimize later. Value would follow.

That's changed. We're starting to see genuine business impact now—not demos or pilots. Real efficiency gains. Measurable productivity improvements. Actual work being displaced by AI taking a more prominent role in workflows. And with that comes a harder question: how do we balance token costs against the people costs we're replacing?

This is no longer theoretical. We have use cases emerging where AI is genuinely changing the economics of work. A process that took three people now takes one person plus AI. A task that took a day now takes two hours. The value is real. But so is the cost of the tokens running at scale.

That's forced a different kind of thinking. It's not "can we use AI here?" anymore. It's "should we use AI here given the token cost, or would a traditional automation be cheaper? Is this a problem AI solves better, or just differently?" The math has to work. The business case has to be clear.

The organizations that are winning right now aren't the ones trying everything. They're the ones being ruthlessly selective—pointing AI at high-value problems where the economics actually work and the impact is measurable. They have examples. Real ones. Use cases they can point to where the trade-off between token spend and people cost is favorable.

Controlling AI spend: token usage is now a real cost line

Here's what nobody quite anticipated: token usage tracking has become critical infrastructure. Not optional. Not "nice to have." Essential.

In the early days, token costs felt like noise—thousands of dollars here or there. Now we're talking about meaningful scale. Run AI across your whole organization at real volume and the numbers get serious fast. Some days the token bill surprises you. Some weeks it's higher than predicted. Some months it's way off forecast.

The challenge is that it's genuinely hard to predict. You can estimate usage. You can set budgets. But then someone discovers a new use case, or a team scales up, or a model becomes more efficient and usage patterns shift. Three months from now the economics could look completely different. A vendor might launch something new. Pricing might shift. The competitive landscape might move.

This means you're making deals differently now. Some tools get enterprise agreements with token caps. Some you use on a unit basis and watch carefully. Some you experiment with and stop. Some you commit to because the value is clear enough to justify the uncertainty.

The intelligent organizations are treating token spend like they treat any other significant cost line—with clear governance, regular reviews, and hard questions about ROI. "We're using a lot of tokens here—what are we getting for it?" If the answer isn't clear, they cut it. If it is clear, they commit and scale.

Fluency still takes time

Sree wrote about how fluency is earned, not delegated. Three months on, that's proven absolutely true. We've watched people who dismissed AI in month one become genuinely fluent by month three because they used it relentlessly. And we've watched early adopters plateau because they stopped experimenting.

The teams that are advancing fastest are the ones treating AI fluency like a skill you actively maintain, not a checkbox you complete. That means regular experimentation. That means trying new tools. That means having conversations about what works and what doesn't. It means accepting that what you know today will be incomplete in three months.

That kind of learning culture doesn't happen accidentally. It needs to be built.

What hasn't changed

The core insights are still holding. Leadership matters more than tooling. Behaviour change is the hard part. Real wins come from pointing AI at your highest-value problems, not your safest ones. The friction is almost always at the junction between technology and how people actually work—and that's where your energy should go.

What's changed is we're getting better at seeing the shape of the work. We're learning faster. We're being more honest about what's working and what isn't.

Moving forward

Across our brands, we're seeing different patterns emerge. Some are moving cautiously—trying, testing, being selective. Others are accelerating. Same tools, same access, but sharper strategy, clearer priorities, better partnership between the business, technology, and the people doing the work.

The brands that are accelerating aren't the ones with the most AI experiments. They're the ones with all three corners of the triangle genuinely moving together. Where the business side knows what it needs. Where technology can deliver it reliably. Where people understand what's changing and why, and have a voice in shaping it.

That's the pattern we're doubling down on. It's not about having the cleverest ideas or the fanciest tools. It's about having the discipline to align all three corners and stay honest about whether it's working.

That's where Travelopia is headed.

What are you seeing three months into your AI journey? Where have your early assumptions held up, and where have they shifted?