The leadership trap
In AI, trying to be the leader is usually the wrong goal. Being an excellent follower, one that reacts faster and smarter than everyone else, is the better strategy for almost every organisation.
Walk into any leadership offsite this year and you will hear the same ambition: “We want to be an AI leader in our industry.” It sounds bold. It looks good on a strategy slide. And for most companies, it quietly sets them up to fail.
Leadership used to mean being ahead of your competitors. In AI, the pace is not set by your competitors. It is set by a handful of frontier labs, OpenAI, Anthropic, Google and a few others, and they play a different game entirely.
Why you cannot out-lead the frontier labs
You are not competing with the labs, but you are measured against them. Every AI initiative you launch will be compared, by your users and your board, to the latest ChatGPT or Claude release. Three forces make that comparison almost impossible to win.
Scale. The labs spend billions on compute, research and talent. Your AI budget, however generous, is a rounding error next to theirs. Any capability you build on your own will be outperformed by a general model sooner than you expect.
Speed. Major model and product updates now arrive every few weeks. A twelve-month AI roadmap is out of date before the steering committee has signed it off. By the time your custom tool goes live, the off-the-shelf version may already do more.
Absorption. The labs keep turning use cases into features. Whole categories of products, from writing assistants to document summarisers to coding helpers, have been squeezed overnight when the same function appeared inside a general assistant. What you call a strategic advantage, they may call next month’s release note.
Try to lead on technology against that, and you end up running fast just to stand still, spending heavily on things that become free.
The case for being an excellent follower
Being first has never been the same as winning. Google was not the first search engine. Apple was not the first to make an MP3 player or a smartphone. Facebook was not the first social network. Each of them watched the pioneers, learned from their mistakes, and then moved decisively. Business researchers even have a name for this: the “fast second” strategy.
In AI, the argument is stronger still. First movers pay the tuition for everyone else. They absorb the failed pilots, the hallucination incidents, the surprise regulatory questions and the tools that never get adopted. A good follower gets those lessons almost for free.
But let me be clear about what following does not mean. It does not mean waiting. A slow follower is simply a laggard with a better excuse. The competitive edge is not in being first or in being cautious. It is in how quickly you can notice a change, judge whether it matters, and act on it.
In a field where the ground moves every month, reaction speed is the strategy.
What good following looks like in practice
Fast following is a capability you build on purpose, not an attitude. From a project and portfolio perspective, it comes down to six habits.
- Sense continuously. Give someone clear ownership of watching what the labs release and what it means for your business. A short weekly briefing beats a big annual trend report.
- Keep your architecture swappable. Avoid locking your processes into one model or vendor. If a better model appears next month, switching should take days, not a new project.
- Pre-approve small bets. Set aside a budget and a lightweight approval path for experiments. If every pilot needs a business case and three committee meetings, you will always be late.
- Run short pilots with clear stop criteria. Two to four weeks, one measurable outcome, and an honest decision at the end: scale it, adjust it, or kill it.
- Build AI fluency across the organisation. The bottleneck is rarely the technology. It is people who do not know how to use it. A fluent workforce adopts new tools in weeks instead of quarters.
- Govern early and lightly. Have your usage policy, data rules and risk checks ready before the next tool arrives, so governance speeds adoption instead of blocking it.
Notice that none of these require you to invent anything. They require you to be organised, disciplined and quick. That is good project management applied to a moving target.
Where you can still lead
Follow on the technology, lead on everything the labs cannot copy. That is where your real advantage lives.
The labs build general tools. They do not know your customers, your processes, your regulatory environment or the twenty years of judgement sitting in your best people. You can lead in:
- Application: how well AI is woven into the way your work actually gets done.
- Data and context: the proprietary knowledge that makes a general model useful in your specific situation.
- Trust: how responsibly and transparently you use AI with clients and employees.
- Adoption: how quickly your people move from curiosity to everyday use.
None of these will appear in a model release note. All of them compound over time.
Rethinking what leadership means
Perhaps the real lesson is that we need a new definition of AI leadership. Not “we got there first”, but “we adapted best”.
The organisations that thrive over the next few years will not be the ones that tried to out-innovate OpenAI or Anthropic. They will be the ones that let the giants run ahead, watched closely, and turned every breakthrough into value faster than anyone else in their market.
In a race you cannot win, the prize goes to whoever adapts fastest.
So here is my question for you: is your organisation trying to lead in AI, or learning to follow brilliantly?