The Last Mile Is Where AI Value Lives
The companies winning with AI aren’t building it. They’re doing the unglamorous work of making it stick.
Yishan Wong, former Reddit CEO and now running a climate tech company, recently issued a blunt warning: most AI application startups are fast becoming obsolete. The foundation model giants, OpenAI, Anthropic, Google, are moving so quickly that the “wrapper” companies built on top of them cannot keep pace. Features that took startups months to build are being shipped natively by the platforms in weeks.
Wong’s argument is about startups, but the implication is broader. If you are a business leader watching AI unfold, the lesson is not “don’t build AI companies.” It is this: the value in AI is shifting. Fast.
From creation to adaptation
For the past two years, the market has rewarded AI creation. The companies raising billions, commanding headlines, and attracting top talent were those building the core models and the flashy applications on top of them.
That is starting to change.
As early as next year, I expect we’ll start to see the market value AI adaptation far more highly than AI creation. The edge will not go to the firms with the cleverest prompts or the shiniest chatbot interface. It will go to the ones who can actually make these tools work inside a real business.
In conversations with leaders already implementing AI, one view comes up repeatedly: the threat is not AI taking jobs. The threat is companies that can adapt AI pulling ahead of those that cannot. That is where the competitive fracture is forming.
The last mile problem
In logistics, the “last mile” is the final leg of delivery. It is the most expensive and failure-prone part of the In logistics, the “last mile” is the final leg of delivery – getting a package from the distribution centre to someone’s door. It is also the most expensive and failure-prone part of the journey. Roughly half of total shipping costs sit in that last stretch.
AI has its own last mile problem. Getting a model to produce impressive demos is relatively easy. Getting it to run reliably, day after day, inside your actual workflows, with your actual data, trusted by your actual people – that is hard. And that is where most AI initiatives quietly stall.
The last mile involves:
Integration. Making AI systems talk to legacy infrastructure that was never designed for them. ERP systems, CRMs, compliance platforms, decades-old databases – all of it.
Reliability. Moving from “works 80% of the time” to “works predictably enough that people will act on it.”
Trust. Getting your teams to actually use the outputs. Not just see them, but change decisions based on them. This is a human problem as much as a technical one.
Governance. Knowing when the model is wrong, who is accountable, and how you catch errors before they compound.
None of this is glamorous. None of it makes for exciting press releases. But it is where the value lives.
The restructuring question
Here is the uncomfortable truth: you cannot solve the last mile by buying more software.
The companies that will gain durable advantage from AI are not the ones with the biggest technology budgets. They are the ones willing to restructure how they work – their processes, their decision rights, their team structures – to absorb these tools properly.
This is not about “AI transformation” as a discrete project. It is about building an organisation that can continuously integrate new AI capabilities as they emerge. Because they will keep emerging, faster than any single implementation roadmap can anticipate.
The firms that treat AI as a procurement exercise – buy tools, plug them in, move on – will find themselves re-buying and re-implementing every eighteen months as the technology shifts beneath them. The firms that treat AI as an organisational capability will compound their advantage over time.
What this means for leaders
If you are running a business or sitting on a board, the strategic question is not “which AI tools should we buy?” It is “how do we become the kind of organisation that can actually use them?”
That means asking harder questions:
Where are we treating AI as a technology problem when it is actually a change management problem?
What workflows would need to change for AI to deliver real impact – and are we willing to change them?
Do our people trust AI outputs enough to act on them? If not, why not?
Who owns the last mile in our organisation? Or is it falling between the cracks?
Are we building organisational capability, or just accumulating software licences?
The bottom line
The AI gold rush has been about creation – building models, launching products, stacking features. The next phase will be about adaptation – the slow, difficult, deeply operational work of making AI actually stick.
That is not a problem you can outsource to a vendor or solve with a flashy pilot. It is a leadership problem. And the companies that recognise it early will be the ones still standing when the dust settles.