Orchestration, Not Expertise, Decides Who Wins with AI
- aiomniconversation
- 7 hours ago
- 5 min read
By Professor Ashley Braganza
Expertise is losing its scarcity value. For decades, careers in large organisations were built on depth: know one thing better than the people around you, and the organisation would find a place for you. That calculation has inverted. Knowledge has become a commodity, and an individual’s knowledge is uncapped by the agents they can mobilise. The professional advice that shaped a generation of managers, specialise early and go deep, may well be the wrong advice. The harder question is what replaces it, and organisations have barely started to answer it.
My guest on The AI Adoption Podcast is Leo Rogers, co-founder and CEO of Curvo, which he describes as the real-time execution layer for sales teams, prompting sellers mid-meeting on questions to ask and objections to overcome. Before Curvo, he built more than one hundred SaaS products in four years for large banks, government clients and high growth startups. He sees adoption from the commercial front line, and his account of it is unsentimental.
Depth has not disappeared, it has moved to the agent
Leo’s formulation of the generalist is more precise than the usual argument for breadth. He is not describing a dilettante. His phrase was that you can be a jack of all trades and a specialist in all of them at the same time, and the heuristic he offers is 30 per cent. Hold roughly 30 per cent of a discipline, enough to set the strategy and to recognise a good output, and let agents supply the rest. Extend that across legal, finance, product and design, and one person can direct work that once required four.
The received insult, a jack of all trades and master at none, is on its way to becoming a compliment. Skill stacking, in Leo’s term, is now the more durable form of job security, because breadth of contribution makes a person harder to remove than depth in a single function.
There is a structural consequence and it is one I put to boards and senior managers I work with. If people are to work usefully alongside agents, they need working knowledge not only of their own function but of the functions their work touches. Agents will push organisations towards cross-functional operation whether or not leaders plan for or want it. That is not a training issue. It is a question of organisation design and, before long, of the business model itself.
Orchestration is the layer that decides whether agentic AI works
Leo described orchestration in operational terms. Generalised agents are slow, because each task sends them back through the same chain of reasoning from the beginning. Specialised agents, trained on narrower data, execute a defined task faster and to a higher standard. Orchestration is the agent that sits above them, takes the strategic instruction from the human, decides which specialists to mobilise, and reconciles what they report back. His verdict is that this makes materially more complex tasks possible.
The point extends. An agent that understands only its own defined task is of limited use if it cannot place that task in relation to the agents around it. The requirement that applies to the generalist employee applies equally to the agent, and orchestration is where that requirement is met or missed. On the evidence of this conversation, orchestration is the layer on which agentic AI succeeds or fails, and it is receiving less board attention than the models themselves. Moreover, whether or not AI technologies are capable of orchestrating to this extent is questionable.
The complication Leo raised should slow any leader down. Judgement of quality still depends on prior exposure. Ask an agent, as a non-designer, to produce a prototype, and you remain unable to say whether the design patterns hold, whether the design system is robust, or whether it fits the brand language. The human gate on quality only works if the human has done the thing before, even manually. An organisation that thins out its junior roles removes the mechanism by which that exposure was ever acquired, and that is a cost that will not appear in the business case.
Laggard organisations and laggard careers fail in the same way
Leo’s profile of the laggard organisation is specific enough to be uncomfortable. It has traded for some decades. It grew steadily on a wave it caught at the right moment, in software that wave being SaaS, and it grew without much risk. Complacency settled into the leadership, and the mandate shifted from growth at any cost to preservation of the shareholders or founders’ family wealth. Private equity then arrives, and in Leo’s account that is where value creation actually begins, because institutional backers have two or three years to squeeze the lemon before flipping the company again. They inherit a laggard technology stack and salespeople ten years into their tenure who are coasting on their black books rather than the CRM.
The individual pattern is the same pattern at a smaller scale, and the parallel is similar. Leo’s observation was that resistance correlates with tenure: one year, three years, five years, and at ten years adoption becomes much more challenging. The reason is identical to the organisational one. The output was delivered and the number was hit, so the input was never scrutinised by leadership. Success removed the pressure to examine the method.
His remedy runs on two tracks, and leaders who fund only the first will get half the result. Organisations build training programmes, and he is clear these cannot be a standard curriculum but one reimagined from the ground up for where their people will have gaps. Individuals build side projects, because those who are not tinkering out of hours will fall behind. He is also honest about the harder version. Retrofitting comes first, but sometimes, in his phrase, a root canal is needed and the organisation has to be restructured.
Human judgement is the one input agents cannot supply
Leo’s account of the human role in an agentic organisation is narrow and clear. Humans foster and create culture, decide what the world should be, and define the vision, mission and products of a company. They set direction, and a fleet of agents executes. His own qualification is worth carrying: this is not the present reality but the direction of travel, and humans remain at the forefront of everything today.
The executive layer survives, in his view, for a reason that has nothing to do with productivity. It survives as a check and balance. His phrasing was that leaving AI to a single madman on his own devices produces a Terminator style outcome. Agents lack empathy, and their judgement is geared to be entirely commercial, which is not always the right frame for a decision. Boards, councils and forums exist to supply the moral compass the commercial frame omits. Middle management, on his account, does not survive that logic.
Robotics follows the same reasoning. The humanoid is a Hollywood inheritance, and its practical uses are constrained. Value is being created by machines shaped by the task rather than by the human body: a bricklaying robot with no legs, sliding up and down a wall at the rate of five bricklayers, set against a humanoid that drops plates and takes half an hour to place a mug in a dishwasher. Agents, purpose built robots and humans working to their respective strengths is the next frontier, and few organisations have designed for it.
The provocation I would leave with leaders is Leo’s inversion of the labour market. Agents will appoint humans for the last mile, because an agent in a sandbox cannot shake a hand or fix a chair. Accept that and the board question changes from which roles to cut to which human capabilities are genuinely irreducible: relationships, strategy, moral judgement, physical presence. Boards that answer that in the next eighteen months will design their organisation. Those that wait will have it designed for them by whoever orchestrates their agents.
Listen to the full conversation
Comments