The Journey to the Frontier Firm

Microsoft's latest Work Trend Index Annual Report sets out a vision of a "Journey to the Frontier Firm" - where organisations in broad terms will move from Phase 1) empowering the individual worker with AI, into Phase 2) hybrid human-agent teams, and then moving towards a Phase 3) of "human-led, agent-operated" organisations.

To make this vision more concrete - consider an example of Agentic AI:
Imagine an AI system where you simply state "Our main competitor just launched a new product line that's gaining traction. Help us develop our response strategy." The AI system asks you for required details, then performs comprehensive market research, analyses your product portfolio, evaluates strategic options, and models different response scenarios - checking back with you (the "human actor") for strategic direction and constraints as needed. The agentic system is able to proceed into suggesting actions across R&D, marketing, and finance teams. Over time this system might learn your company's strategic preferences and risk tolerance, performing more of these analyses with less guidance and more autonomous recommendations.
Now imagine the same capabilities - but applied to your business domain, and imagine layers of such systems building upon one another across the enterprise landscape - the output of one system, becoming the input for the next - Agentic AI Systems interaction with each other. Then imagine the capabilities of the underlying AI models continuing to improve - and you have the basis for Microsoft's vision of the the "human-led, agent operated" Frontier Firm. 

Companies like Deloitte have described similar structures calling them Superteams - humans setting strategic direction - whilst AI agents execute business processes, checking in as needed. It's a compelling narrative of efficiency, scale, and superhuman capability. 

For leaders who believe they understand their market and their competitive landscape, this vision taps into something fundamentally seductive: the ability to codify their expertise, multiply their decision-making capacity, and execute their strategy with unprecedented precision. It promises to transform the messy, unpredictable world of human organisations into something more like a well-oiled machine.

But - there's a potential problem lurking beneath this glossy surface. While the short term efficiency gains of Agentic AI systems can seem imperative - you might want to consider the impact on your long term competitiveness. 
As Mark Twain told us many years before the first computer was ever built: 

"It ain't what you don't know that gets you into trouble.
It's what you know for sure that just ain't so."
 – Mark Twain 

The Hierarchical Assumption Lock-in

The potential risk in the Agentic AI vision lies not in the technology itself, but in how it interacts with human organisational behaviour. When leaders - out of the best intentions - believe that they "know something for sure", they are naturally inclined to select subject matter experts (SMEs) who align with that worldview. They might then decide to empower these SMEs with Agentic AI systems that can greatly increase their productivity, removing the need for junior staff and lots of expensive training. These powerful Agentic AI systems, might in turn also be built under the direction of the same leadership.
The result is what we might call "Hierarchical Assumption Lock-in" - a systematic reinforcement of potentially false assumptions throughout the organisation.

This phenomenon isn't new. Political theorists have long warned about technocracy - rule by experts - precisely because experts can share the same blind spots - and they might prefer other experts who share the same expert view, entrenching rather than challenging the status quo.  
In the context of Agentic AI, this problem becomes exponentially more dangerous. Unlike human colleagues who might challenge your assumptions (even if awkwardly or inconveniently), AI systems will only argue with you if you instruct them to do so. And here lies one of many traps: how can you correctly instruct an AI system on when and how to challenge your assumptions? How would you know where your own fundamental premise might be wrong, and should perhaps be challenged? 

The seductive efficiency of the "human-led, agent-operated" Frontier Firm - carries an inherent potential for closed feedback loops. Leadership is inclined to pick SMEs who align with their own world view. These SMEs are empowered with AI agents that are in turn created by the same organisation with the same fundamental assumptions and world view. 
Where, in this elegant system, is the mechanism that challenges the fundamental assumptions upon which the entire edifice is built? 

The Innovation Paradox

Take Steve Jobs and the iPhone as an extreme but illustrative example. Would an Agentic AI system, designed by the same management mindset that created the Apple Newton, have conceived of the iPhone? The iPhone wasn't the result of incremental improvements to existing technology - it was a fundamental reimagining of what a mobile device could be. It required the kind of quantum leap that human cognition specialises in: jumping from very few facts to a radically different, possible future state.
Consider for a moment the countless breakthroughs in human history that came not from the systematic execution of expert knowledge, but from unexpected insights that challenged conventional wisdom. These moments of genuine innovation share a common characteristic: they emerge from the collision of different perspectives, the questioning of established norms, and the willingness to say "but what if we're completely wrong about this?"

This is where the current rush towards Agentic AI becomes particularly concerning. While AI excels at pattern recognition, data processing, and systematic analysis, human cognition offers something entirely different: the ability to combine imagination, drive, empathy, and ingenuity in ways that transcend logical progression. 
When we replace human actors with AI agents, we don't just lose efficiency gains - we lose the very source of breakthrough innovation - and that means endangering your long term competitive advantage!

The Long-term Competitive Threat

Organisations rushing towards fully agentic systems may indeed reap short-term benefits: faster execution, more consistent outputs, reduced operational costs. However they risk something far more valuable: their capacity for genuine surprise and breakthrough innovation. In a world where every organisation has access to similar AI capabilities, the differentiating factor won't be who can execute existing strategies most efficiently - it will be who can discover entirely new strategies that others haven't considered.

This is where maverick human agency becomes not just valuable, but essential. 
Most leaders have probably experienced joy and pride when a task delegated to a junior colleague resulted in that colleague coming back with a genuinely novel, ingenious solution or opportunity discovered.
These interactions cannot be programmed because, by definition, they challenge the very assumptions upon which the programming is based. Subject matter experts, reinforced by their organisationally validated assumptions - might over time become too entrenched in that thinking. This is where you need mavericks and free thinkers, whatever you may think of them - your organisation needs them!

The current job market trends - with graduate employment at its worst levels in over 40 years, partly due to AI replacing entry-level positions - represent more than just economic disruption. They represent the systematic elimination of the very pipeline that has historically produced organisational mavericks. When companies use AI to handle tasks traditionally given to junior staff, they don't just save costs - they lose the fresh perspectives and challenging questions that often lead to breakthrough innovations.

The Case for Maverick Human Agency

The solution isn't to reject AI, but to fundamentally reconsider how we deploy it. Rather than rushing into human-led, agent-operated systems, there's compelling value in exploring human-empowered, AI-augmented approaches that preserve what makes organisations most competitive: their capacity for genuine surprise.

The most successful implementations we've observed tend to share three characteristics: 
1) They focus on empowering human actors with AI - rather than replacing them.
2) They maintain meaningful human oversight in decision-making processes, with AI providing analysis and options - rather than autonomous choices. 
3) They use AI to enhance human cognition - rather than attempting to substitute it.

The competitive advantage in the AI age may well come - not from the most sophisticated automated systems - but from the most creative, empowered, and diverse human teams augmented by thoughtful AI deployment. Whilst competitors build closed-loop systems that inadvertently reinforce existing assumptions, forward-thinking organisations might instead create open systems that continuously challenge and evolve their thinking.

The future likely belongs to organisations wise enough to preserve and empower maverick human agency. The junior colleague who returns with an unexpected solution. The team member who challenges fundamental premises. In a world of increasingly sophisticated automated systems, the scarcity might prove to be the uniquely human ability to imagine what doesn't yet exist and to question what everyone else takes for granted.

Whether you like a maverick or not - your organisation needs them - and you have to accept that you might not be able to identify and pick them - you might even disagree with them.
Because what you know for sure, sometimes, it just ain't so!