Most organisations today invest heavily in AI tools. Few invest in the ability to orchestrate human and machine intelligence as a coordinated capability. Yet research suggests that the real leverage lies precisely in this orchestration: in consolidating data foundations, leading mixed teams of people and AI agents, and governing adoption at the enterprise level. This article draws on recent studies from Harvard Business School, Salesforce, Gartner, and others to outline why fragmented AI adoption falls short, what a human-centric approach to AI orchestration looks like in practice, and how organisations can move from isolated experiments to measurable returns.

The Starting Point: Fragmentation Instead of Integration

AI is reshaping how organisations create value, and doing so faster than many can respond. That much is well understood. What is less often discussed is that AI adoption, as it plays out in most companies today, produces more fragmentation rather than less.

Two problems drive this. The first is fragmented data. Teams want to use AI, but the relevant information sits scattered across disconnected systems: CRM here, ERP there, industry databases elsewhere. Without unified access to this data, AI cannot deliver reliable outputs.¹ The second is uncoordinated intelligence, both human and machine. Individual employees and teams develop their own isolated solutions: separate subscriptions, standalone agents, no shared knowledge base. This may work at the level of a single user, but it does not scale, and it generates no measurable impact at the organisational level.

The Goal: A Long-Term Return on AI Investment

If AI adoption is done well, it delivers a long-term return on investment across three levels. First, relief: AI frees teams and leaders from repetitive tasks, creating space for work that requires creativity and judgement. Second, acceleration: when deployed efficiently and under oversight, AI optimises processes and reduces resource expenditure. Third, and this is the most consequential level, innovation: AI enables entirely new forms of value creation, not through automation alone, but through human-AI collaboration that makes previously impossible outcomes achievable.³

These three levels form the target picture. But two major obstacles stand in the way: data management and leadership.

The Data Problem: Garbage In, Garbage Out

The principle is straightforward but routinely ignored: the quality of any AI output is determined by the quality of its input. No matter how sophisticated the model or how large the dataset, if the underlying data is flawed, the results will be too. This is not a technology problem. It is a data-foundation problem.¹

And it is not a marginal issue. A Salesforce study found that 81 per cent of organisations struggle with AI adoption precisely because they cannot consolidate and cleanly integrate their data.² Eighty-one per cent. The data hurdle is not one challenge among many. It is the central challenge.

In a world where markets, technologies, and regulatory requirements shift rapidly, organisations need flexible data foundations: not only structured internal data deposited once and never revisited, but dynamic integrations that incorporate external signals such as competitor intelligence, market developments, and regulatory changes. The analogy to human decision-making is apt: when professionals make judgements in their daily work, they do not rely on a single data source. They integrate what they have read, heard, and experienced. An AI's data foundation needs to function the same way.²

From a Technical Problem to a Leadership Problem

This agility cannot be solved through technology alone. Who decides which data sources are integrated? Who is responsible for quality? Who ensures the data foundation stays current? AI is not simply a software rollout. It is a reorganisation of work that requires central governance with clear roles, clear responsibilities, and clear decision rights.⁴ The technical data problem, in other words, leads directly to a leadership problem.

When the conversation turns to leadership and AI, the familiar question surfaces: does AI replace jobs? The reality is more nuanced. AI does not replace jobs. It augments teams. Roles become more complex, not fewer. And, critically, they increasingly require the ability to collaborate with AI. Workday articulated this well in a recent announcement: "As digital workers increasingly support workforces, organisations must rethink how they hire, manage, ensure compliance, and measure their impact, just as they would for human employees."⁵ The implication is significant: organisations must lead, steer, and manage AI with the same rigour they apply to human employees. For any leadership team, this represents a fundamental shift.

Teams With AI Outperform Individuals With AI

But how exactly should this collaboration be structured? Is it more effective for each individual to have their own AI assistant, or for entire teams to work with AI collectively?

A Harvard Business School study analysed nearly 800 product developers at Procter & Gamble and produced clear findings.⁶ ⁷ In terms of qualitative output, individuals working without AI delivered inconsistent results. Teams without AI performed notably better. Individuals with AI assistants improved further still. But teams collaborating with AI delivered the highest qualitative output overall.

Some may argue that it is nevertheless more efficient to equip each individual with their own AI agents, following the so-called "agent boss" model, where one person steers a team of AI agents. And in terms of time savings, the data does bear this out: individuals with AI are faster than teams with AI.⁶ So why not simply scale this model?

Because the top results come from elsewhere. When the researchers examined the best ten per cent of all outcomes, the genuinely outstanding innovations, these came from teams working with AI, not from individuals with AI.⁶ ⁷ The strongest results emerge when people and AI collaborate as a team.

There is another dimension that is often underestimated: the teams that worked with AI also reported the highest levels of satisfaction and the most positive emotions about their work.⁶ Human-AI collaboration is not only more effective. It is experienced positively by the people involved. That matters considerably when the objective is sustained adoption.

The Shadow AI Problem

So, the encouraging news is that employees want to work with AI. The less encouraging news is that much of this collaboration is already happening outside any organisational framework.

A study of more than 600 leading IT companies found that employees use AI on their own initiative, with significant risks for output quality, data protection, compliance, and cyber security. It is notable that even among IT firms this problem is so pronounced. At 61 per cent of the companies examined, unauthorised AI tools were found in use among staff.⁸ At the same time, only 26 per cent of organisations had any solutions in place to monitor AI usage.⁹ And nearly half, 49 per cent, reviewed AI usage within their organisation either not at all or only reactively, after something had already gone wrong.⁸ That is not a strategy. That is risk management in the rear-view mirror.

Human in the Lead, Not Human in the Loop

This is why the conversation must shift to leadership. AI agents, no matter how capable they become, are not "set and forget." They are learning teammates that require human oversight. The technology is not yet at a point where agents operate fully autonomously and without error. Ritcha Ranjan, Senior Vice President at Expedia, captures this precisely: "As we move towards more agentic workflows, that human-in-the-loop moment is going to be critical."¹² ¹⁰

But there is an important distinction. It is not enough to simply keep the human "in the loop." For human-machine teams to function well, the human must serve as the central control authority. Julie Sweet, CEO of Accenture, put it concisely at Davos: "Human in the lead, not human in the loop."¹¹ Not merely involved, but in charge.

From this understanding, a new leadership task emerges. When organisations operate heterogeneous teams of people and AI agents, when collaboration requires coordination, when governance and compliance must be ensured, what is needed is a holistic, human-centric orchestration of total intelligence across the enterprise. AI is no longer a purely IT concern. This orchestration demands stewardship at the highest leadership level, from the CIO to the CHRO.

A Six-Point Plan for Human-Centric Intelligence Orchestration

Based on current research and practical experience with organisations, six steps can keep the human in a leadership position while unlocking AI's full potential within agile teams.

1. Consolidate the data foundation before scaling. Before introducing new tools or investing in expensive AI models, organisations need to unify access to internal data and external market signals. With 81 per cent of companies struggling at precisely this hurdle, the data foundation is the prerequisite for everything that follows.²

2. Create a secure environment instead of isolated solutions. Employees are already using AI, often on their own, with all the associated risks. Rather than prohibiting this, organisations can provide their teams with a flexible, controlled environment where they can experiment, build workflows, and learn from one another, without sensitive data leaking or security risks materialising. This turns shadow AI into governed innovation.⁸ ⁹

3. Start with routine tasks and build trust. Begin with the "no-joy" work, the repetitive tasks that nobody enjoys. Quick efficiency gains build trust within teams. And trust opens the door to the more complex collaboration required for genuine innovation.³

4. Position AI as a teammate, not a tool. AI can serve as a market analyst that monitors external signals and alerts the team when developments require a response. It can function as a resource planner preparing internal data for team coordination. As a strategy manager maintaining oversight and prioritising projects. Or as a compliance manager identifying and mitigating risks. These are not future visions. They are use cases that work today. The critical point: these AI teammates need clearly defined roles and responsibilities, just like human team members.⁶ ⁷

5. Define clear rules of engagement. Who decides what? Organisations need working agreements for human-AI collaboration: when does AI deliver suggestions, when do humans decide, and how are results reviewed? Clear roles, responsibilities, and decision rights are essential. Without rules of engagement, there is no collaboration, only disorder.⁴

6. Measure adoption and iterate. What cannot be measured cannot be steered, or rewarded. Organisations should establish monitoring for AI usage, not only for control but above all to make successes visible. When it becomes clear which teams practise effective human-AI collaboration, others can learn from them. This creates a positive feedback loop: measure, make visible, iterate. What works today becomes standard tomorrow, and the day after requires the next step.⁹

Orchestration Needs Platforms

Implementing these six points, the monitoring, the secure environment, the governance, is not a purely human task. AI supports the human here as well. It is the same pattern at every level: AI becomes integrated into leadership and governance itself. This means organisations need not only new leadership principles but AI-supported systems that enable central human stewardship.

These are known as orchestration or governance platforms. That this is not a niche concern is demonstrated by Gartner's findings: a survey of 360 companies showed that organisations with AI governance platforms were 3.4 times more likely to steer AI successfully. Gartner expects such platforms to reduce regulatory costs by up to 20 per cent, freeing budget for growth. The investment pays for itself directly, not only through better governance but also financially.¹³

How BlackMountain Approaches Orchestration

BlackMountain builds a platform that operates across three layers. First, consolidated data: all external signals and internal organisational knowledge unified into a single source of truth, hosted on European servers and fully encrypted. Second, orchestrated AI: all AI models, agents, and tools, from GPT, Claude, and Gemini to Mistral and custom models, centrally coordinated in one interface, with full transparency over usage, costs, and compliance. Third, complete context: every conversation, every signal, every document remains structured, retrievable, and compliance-conformant, stored in an organisational memory that turns each interaction into a lasting asset. The result is a working environment where the human judges and AI supports, and where teams choose how they work and what they prioritise, backed by the right tools and assistants in a single, secure interface.

Closing: A Leadership Principle, Not a Technology Decision

Human Centric Intelligence Orchestration is, at its core, a leadership principle rather than a technology decision. The research is consistent: the strongest outcomes emerge when people and AI work as coordinated teams, when data foundations are consolidated rather than fragmented, and when governance is proactive rather than reactive. Relief, acceleration, and innovation, the three levels of long-term AI return, are not achieved by adding more tools. They are achieved by keeping the human at the centre and orchestrating AI around human work, not the other way around.

——

Sources

¹ Rand, B. (2025). Bad Data, Bad Results. *Harvard Business School Library*, 27 February 2025.

² Integration Key as 97% of APAC IT Leaders Turn to AI Agents. *Salesforce News*, 6 February 2025.

³ Chen, W.X., Srinivasan, S. & Zakerinia, S. (2025). Displacement or Complementarity? The Labor Market Impact of Generative AI. *Working Paper*, Hong Kong University of Science and Technology / Harvard Business School, 2025.

⁴ Layne, R. (2025). AI Trends for 2026: Building 'Change Fitness' and Balancing Trade-Offs. *Harvard Business School Library*, 18 December 2025.

⁵ Workday Announces New AI Agent Partner Network and Agent Gateway to Power the Next Generation of Human and Digital Workforces. *Newsrelease*, 3 June 2025.

⁶ Dell'Acqua, F., Ayoubi, C., Lifshitz, H., Sadun, R., Mollick, E., Mollick, L., Han, Y., Goldman, J., Nair, H., Taub, S. & Lakhani, K.R. The Cybernetic Teammate: A Field Experiment on Generative AI Reshaping Teamwork and Expertise. *Harvard Business School Working Paper*.

⁷ Nover, S. (2025). When AI Joins the Team, Better Ideas Surface. *Harvard Business School Library*, 17 October 2025.

⁸ Shadow AI Governance Lags as AI Adoption Soars, According to Cato Networks Survey. *Cato Networks Newsroom*, 3 December 2025.

⁹ Sanchez, A. (2025). New WalkMe Survey Shows Shadow AI Is Rampant; Training Gaps Undermine AI ROI. *SAP Newsroom*, 27 August 2025.

¹⁰ Blanding, M. (2026). What Leadership Looks Like in an Agentic AI World. *Harvard Business School Library*, 11 February 2026.

¹¹ Hoang, J. & McCallion, P. (2026). Davos 2026: Leaders on why scaling AI still feels hard — and what to do about it. *Weforum.org*, 23 January 2026.

¹² Neeley, T. & Ranjan, R. (2025). Generative and Agentic AI as Strategic Partners for Leaders. *Harvard Business School Technical Note 426-038*, November 2025.

¹³ Global AI Regulations Fuel Billion-Dollar Market for AI Governance Platforms. *Gartner*, Stamford, Conn., 17 February 2026.