Adoption vs Adaptation
Constellation Research held their annual Connected Enterprise event last week, and unsurprisingly, enterprise AI was very much top of mind for many of the CIOs and analysts in attendance. But as Jon Reed reported for Diginomica, there are signs of a widening gap between what enterprises are concerned about and the rhetoric and messaging of the software vendors.
Enterprise IT has a tough job right now in satisfying senior stakeholders by ‘doing something’ in AI, whilst also trying to build out long-term foundations for success and trying to balance this with security, privacy and workforce considerations. We sometimes say that technology is the easy part, compared to organisational change, but in a legacy environment with multiple geological layers of systems and plumbing, it is never really that easy.
Are other corporate functions currently pulling their weight?
For the moment, at least, it seems that too much of the enterprise AI adoption burden is being borne by technical teams and departments, rather than leadership as a whole engaging with the complexity of the challenge.
As Alex Worapol Pongpech put it last week:
Everyone loves to say they’re building an Enterprise AI strategy. But if you ask five people in the same company — say, a CEO, a CFO, a CTO, a Head of Data, and a middle manager — what “Enterprise AI” means, you’ll get five different answers.
The legacy of functional divisions and competing C-suite roles in the Twentieth-Century corporate model can make it hard to get things done that require cross-functional collaboration, platforms and connected systems. And if leaders don’t step up to solve these organisational limitations, the default position is often just to mandate “action” from above.
We have seen this story play out before and it often leads to problems such as:
- Buying licenses (aka ‘doing something’) ahead of knowing what to do with them
- Disappointing ‘adoption’ based on addressing the easiest BAU use cases
- Long-term over-reliance on the very SaaS vendors that have held them back for so long
Traditional enterprise IT vendors like Oracle and Salesforce will be happy to cure this headache, and are adding new AI capabilities to their legacy platforms so that they can turn painful complexity and opportunity into a simple IT purchasing decision:
Salesforce’s plan to acquire Informatica will unleash a powerful trifecta of technologies, making it easier for organizations to benefit from a new, human/digital hybrid workforce
Or perhaps companies will be persuaded to hand over their closely-guarded data to OpenAI or their competitors, who are exploring ways to address the enterprise market with their own platforms and models:
On October 23, 2025, OpenAI launched “Company Knowledge” for ChatGPT Business, Enterprise, and Edu plans — a feature that searches across internal tools like Slack, Google Drive, SharePoint, GitHub, and Gmail to generate responses with citations. OpenAI COO Brad Lightcap stated on X that “company knowledge has changed how I use ChatGPT at work more than anything we have built so far.”
This would echo what happened with enterprise social networking and knowledge sharing around the mid-2010s, when many companies decided it was too hard to build their own digital business capabilities, and it was easier to allow Facebook inside the firewall with its now-discontinued Workplace by Meta (née Facebook at Work).
Interestingly, it seems there are internal fissures developing within OpenAI itself between AI purists and the large number of Meta staff who have been tempted across, and who have brought Facebook’s customer engagement ‘philosophy’ with them. You don’t need to be an alarmist to foresee the potential harms of AI + advertising + hyper-personalisation à la Facebook as a revenue strategy for OpenAI.
These struggles to embrace AI tools and the change they make possible within our organisations are not a surprise, but we have to hope that leaders can come together and do the work around AI readiness and world-building to enable AI-led transformation of our organisations into smart, connected, human-led engines of value creation. If not, and if they are not up to the task, then they will be renters of big tech good-enough platforms, rather than builders and owners of the future. The old terminology of adoption and change management rather than adaptation and change acceleration might be part of the problem here. We need fresh thinking.
The potential benefits of doing this work are too good to ignore.
I was a guest on the Boundaryless podcast recently, where I tried to make the optimistic case for enterprise AI being a game changer for how we organise work and value creation in the modern economy. We had a great conversation that connected AI with organisational architecture, the role of leaders and AI’s relationship to the platform organisation model:
The Wrong Kind of Circular Economy
But fears about big tech potentially misusing AI are not just about platform dominance or consumer protection – they are also about efficiency and innovation.
A risk for both consumer and enterprise AI sectors alike is that we bake old inefficiencies and outdated ways of working into the foundational layers of what could be transformative business technologies.
In the consumer AI space, there are growing fears of the AI bubble bursting – not the tech bubble that hyped up the tools, but the financial bubble that surrounds AI investment.
In the context of the current geo-political tussle between the United States and China, the US strategy of throwing infinite money and compute at brute force scaling has led to a weird circular economy between chip makers, LLM providers and other players such as hyperscalers and data centre builders. It is also likely to be a highly inefficient use of capital.
Meanwhile, China seems focused on generating efficiencies at every level, from power consumption to model reasoning, and pursuing open models that could enable a far wider opportunity for app builders and practical industrial or enterprise applications. Their total addressable market will be far bigger, more competitive, but also with greater diffusion effects, whereas the US AI market looks more like a casino with a few high rollers battling to win the whole pot.
Right now, the Financial Times is full of commentators and commenters who fear the implications of this bubble bursting and the expected stock market crash that might follow, since it is really only the AI bubble that is driving valuation growth in the United States.
As AI doomster Gary Marcus points out (and a few FT commentators also realise), a paradoxical effect of trying to limit China’s access to the latest AI technologies could be that China commoditises open models that are as good as leading LLMs, and does so without relying on Nvidia chips, which could prick the bubble of ever-increasing big-7 AI valuations.
AI tech is real, not just hype, regardless of financial markets. And big tech earnings are also real, despite their inflated valuations. So this is not really the same story as the dotcom crash. Just as Amazon survived that crash, so will substantive AI tools, technologies and at least some of the leading firms, if it happens again. And for enterprises and consumers alike, we might later thank investors for paying for a huge amount of infrastructure that will enable more sustainable innovation after the crash, just as happened with telecoms infrastructure post-dotcom crash.
Renting Faster Horses vs Buying a Car
But perhaps our financial models and incentives are also holding us back in other ways, and keeping too many firms focused on faster horses rather than new ways of doing things?
In the consumer AI space, you can use AI models that cost billions of dollars for $20 per month and use them to do your homework or make a PowerPoint deck. In the enterprise space, the models cost a lot more, and companies typically buy licenses in batches of thousands (volume discounts!), often before they have a clear idea of what people might use them for.
The quick hit of using somebody else’s GenAI model to perform a task or provide an answer is addictive and feels like magic. Having access to LLMs in this way is great and we can expect this to be a permanent feature of enterprise technology that improves along the way as we adapt generic models to the specific context of each organisation.
But until we put at least as much effort, forethought and investment into organisational readiness and adaptation, these kinds of marginal efficiency gains will be as good as it gets, despite barely scratching the surface of what is possible within an organisational context.
The real prize is smarter, simpler organisations that orchestrate and automate workflows and basic back-of-house functions to give people and teams powerful business capabilities they can use to create value for their customers and colleagues. This could remove the need for messy bureaucracy and costly administrative management layers, and make companies more agile, cheaper to run and easier to scale up.
And returning the point about fragmented leadership not doing enough to help enterprise IT make sense of AI strategy and organisational transformation, it is entirely possible to tackle this challenge with step-by-step transition strategies where each new state of operations shows cost savings and benefits on the way to a more automated, programmable organisational operating system.
This is where we need to focus. People are quite capable of discovering clever use cases for individual prompting if we allow them to learn from each other and share in the open. But they can’t change the system they work in without leadership being aligned on medium- to long-term goals for Agentic and Enterprise AI, and doing the work to transform the organisation.
Last week, we wrote about our emerging framework for AI world-building in the enterprise that can bring together the various levels of learning and transformation goals from the system level to the workplace experience level. Enterprise IT functions have a lot to do at the system level to connect data, services and systems to create a technical foundation for the AI-enabled organisational operating system. But leaders and business functions need to roll up their sleeves as well and come together around an organisational strategy, not just a technology roadmap.