Kathy Gibson reports – Agents are the hot topic of the moment, with the potential to propel artificial intelligence (AI) to new heights as they are used to automate tasks.
“This is the biggest thing I have experienced in 30 years as a consultant,” comments Sylvain Duranton, global leader of BCG X.
A new survey from Boston Consulting Group (BCG) and MIT Sloan Management Review (MIT SMR), confirms that agentic AI is rapidly entering the enterprise, with 35% of organisations already starting to use it and an additional 44% planning to so soon.
Duranton explains that, unlike traditional techs, agentic AI combines the qualities of a tool and a teammate. The survey found that 76% of respondents now view it more like a coworker than as a tool.
This dual nature challenges traditional management approaches, creating deep tensions that call for new solutions across organizational design, investment, governance and process management.
This has created something of a paradox, Duranton adds: organisations recognise that agents make a difference – 73% say it increases the ability to differentiate organisations from each other, and 73% says it allows individual to differentiate themselves from coworkers – but almost half (47%) of organisations still lack an AI strategy.
Despite a lack of strategies, Ai adoption is accelerating. Adoption of machine learning and generative AI (GenAI) continue to rise, while 35% of organisations are already using agentic AI, with another 44% planning to do so soon.
The main reason for its adoption, Duranton says, is that it allows for differentiation.
The way workers relate to agentic AI is also different from earlier technologies, and it is more like to function like a human worker (76% of respondents) than simply as a tool (24%). In Africa this gap is even wider, with 82% of respondents considering it more like a colleague (82%) than a tool (18%).
The rise of agentic AI raises new challenges for management:
- Scalability versus adaptability – agentic AI combines tool-like predictability and human like flexibility;
- Experience versus expedient – the need to balance long-term capability with short-term returns;
- Supervision versus autonomy – traditional oversight systems with full human control or complete automation, or a range of blended models;
- Retrofit versus reengineer – retrofitting AI into existing workflows can drive quick incremental gains, while reengineering processes has the potential for transformative but slower refits.
Nicolas de Bellefonds, global leader of BCG’s AI and software business, outlines the five learnings that emerged from the study. They are:
- Anchor decision-making in value;
- Redesign work around agentic-first workflows;
- Upgrade governance to specifically address agents;
- Redefine agentic-human hybrid roles; and
- Develop learning loops for agents across their lifecycle
He points out that 66% of leading organisations expect a change in operating model as they deploy agentic AI.
With AI rapidly growing in its ability to perform key elements of today’s jobs, organisational processes need to change. Thirty-four percent of respondents say this must happen now, and 49% say the change will come within three years. African companies are more ambitious, with 47% expecting changes now and 60% in the future.
Meanwhile, 58% of leading organisations expect a change in governance and decision-making rights.
The survey determined that, as AI systems increasingly gain autonomy, leaders must develop adaptable governance structures to address:
- AI systems work that work independently from humans –14% now, 39% in three years;
- AI systems that have decision making authority – 19% now, 35% in three years; and
- AI system working with ambiguous inputs – 37% now, 57% in three years.
Agentic AI can take on multiple roles, so organisations need to change how they think about the structuring around them. They are expected to functions as:
- Assistant – 25% now, 63% in three years
- Colleague – 11% now, 35% later
- Rival – 8% now, 25% later
- Boss – 5% now, 18% later
- Mentor – 12% now, 36% later
- Coach – 13% now, 42% later
There were some surprises in the types of jobs that could be replaced by agentic AI, leading to flatter organizational structures.
Within three years, 29% of respondents expect a reduction in entry-level employees, 43% believe they will hire fewer specialists and more generalists, and 45% expect to see a reduction in layers in middle management.
“This is the biggest impact expected across the board,” Bellefonds says. “The general acceptance that juniors will be most impacted is wrong. This implication could be much more disruptive.”
AI agents need full lifecycle support, just like human resources for humans. And the humans in the organisation will also have to learn how to work with agents.
Agents need to be recruited to develop and evaluate agent, onboarded to test and validate, training and retrained to finetune them as more data becomes available, evaluated to track accuracy, adaptability and biases, and retired as they become obsolete.
“So organisations also have to upskill humans to orchestrate agents, not just operate then. They need to supervise, redirect and critique agent outputs – not just consume them,” Bellefonds concludes.