Automation is a fraction of the prize. Refound the business, do not pave the cow paths. The hard part is always people, never the technology.

The company name draws a double meaning. The CEO does not like naming companies, so he left it to a co-founder, but the deeper reason is that kung fu means anything that takes discipline to master, not just the martial arts. Kung fu around AI reflects how hard it is and how important it is to do it right. The purpose is to help clients into the AI age and help them succeed, reinventing themselves to develop and sustain unique advantage from the application of AI.

The current trend towards automating everything to reduce costs and create efficiencies is misplaced. It fails to take the full extent of the opportunity into account, and it treats substituting AI for people as the best path to value. Automation has its times and places, but it is a tiny piece of the potential.

Michael Hammer said, do not pave the cow paths. This is an opportunity to think, where should the road go, not where have the cows been for 100 years. Refounding means going back to what made your company great and got you to this point, refocusing on those principles, and letting go of the assumptions and functional designs that are no longer necessary. It is a renaissance for leaders to re-envision what their company can look like in this new era.

Consultancies are acutely feeling the pain. Many chased the high margin opportunity, which for years was enterprise software and the services around it. AI is threatening that premise, so the question becomes whether you need people in seats to facilitate an installation, or whether agents can do it. KUNGFU.AI is structured differently, not as staff augmentation or substitutes for employees, but bringing something net new, either technical performance a client could not reach alone or a roadmap to transformation they cannot generate on their own.

So much of what wrapped around enterprise software was customisation and data linkages, and a lot of that is now self-service. You can set up the linkages yourself with an agent or a customisation engine, tell it exactly what your use case is, and avoid the painful process of turning off half the functionality in a product you just bought so it works in your context.

We do not do data plumbing if all a client wants is a warehouse or a lake. We do it when it is critical to delivering the value of the AI. One company had proprietary data in 20 different locations across continents, and that data could never leave. We built a federated learning system across all those repositories and trained the model as if it were in one common place. Without tackling that engineering challenge, the model's value could not have been delivered.

That was healthcare diagnostics, where most data lived in hospitals, could not leave, and carried highly protected personal information. The firm works quite a bit in healthcare and life sciences, in finance and fintech, and in government, across security work and the IRS, always mindful of the regulations and unique requirements.

The mistake is thinking you can get your data in one place, put a foundation model in front of it, and value falls from the sky. Most data is messy and not built for the purpose, so without curation you get out what you put in, and you can inadvertently perpetuate a failed project or a loss by telling the AI it is good to access. Treating it like a coworker is a mistake, because coworkers exercise judgment and leverage prior experience and shared history, and the AI does not do that out of the gate.

Even if you interlink data across silos, if it is not curated for a common purpose you hit the same problem. Attribution is notoriously difficult, and everybody thinks connecting data across silos will solve it. But if collection cadences do not match, or you lack the metadata on origination, you either cannot make the attributions you want or you make false ones, which is worse.

Giving everyone unfettered access to foundation models and hoping value falls from the sky rarely works, and when it does it is in pockets you cannot attribute spend to. You create a Wild West. More problematic is failing to set up governance, which does not mean just controls but training people on what good looks like. Without it, agents surface salary data or expose personal information. Nobody sets out to release that data, but the AI does not know it is not supposed to take advantage of it.

Too much governance, or governance focused only on preventing mistakes, drives people to work around you or bring their own licence. We have had senior C-suite leaders putting proprietary company data on a public Claude licence because they lacked the most basic understanding of privacy requirements. The fix is to socialise policies, make sure they are understood, and enable people, turning governance from a control mechanism into a value creation mechanism.

Compliance offices are built to restrict more than to enable, yet AI touches every part of the organisation, so it needs more than restrictive covenants. Give a CTO the full mandate and you hit a different scope and a different budgeting mechanism, when that office was historically a cost centre, and governing human behaviour is not the same as governing security. Put it in a chief product office and the focus becomes shipping features fast, which creates downstream complaints from users who do not understand the technology.

Right-size the approach to where the organisation is in its journey. Companies early on do not need heavy restrictive policies, but they do need people focused on financial accountability so they do not lose money on a bad investment. Large enterprises with thousands of projects in flight must run the full gamut of governance and enabling, because their future depends on people being able to adopt and innovate. The whole company owns the AI, just as the whole company owns the software it sells.

This is not an easy thing to set out to do. It is centred in changing the basis of your organisational structure and how people think about how value is created. It is not about the technology, it is about the people, and executing change across a large base with long tenure and a strong culture is hard. You get early momentum because the CEO is on board, but if the CEO loses focus or declares victory, things regress and all those gains are lost.

Leaders do a lot of self-driven research and arrive saying they asked Claude or ChatGPT. It rarely oversimplifies things for them. More often it opens their eyes that this is harder than they thought, which is a good entrance, because the notion that it should be easy, cheap or short has already been dispelled.

A great asset is access to people outside the strategy and organisational change space. The firm's engineering staff keep him current on technical change, and in regular conversations they unpack what each change means for organisations. On his own he reads a lot, carefully, because AI has proliferated a lot of slop, following a few people who think the right way about it rather than any single source.

It mirrors how companies should think about human machine integration. Focus on the things the human does well, and ask what you bring to the table that is difficult or impossible to automate or replicate with AI. Then understand how to work with the AI so the aggregate output of your collaboration is better than the sum of the parts. One plus one equals three.