For much of the past two years, many business conversations around AI began the same way.
Everyone was obsessed with “What’s the best prompt for this?”.
And many sessions were dedicated to learning how to phrase better questions, prompt frameworks and building prompt libraries.
That phase was critical to help organisations become comfortable working with AI.
But prompting, for all its usefulness, didn’t change how work actually moved.
While people became faster, business structures stayed the same
The real shift only came later when AI stopped waiting to be asked and began carrying work forward on its own – That was until the emergence of 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗔𝗜.
𝗙𝗿𝗼𝗺 𝗣𝗿𝗼𝗺𝗽𝘁𝘀 𝘁𝗼 𝗔𝗴𝗲𝗻𝘁𝘀: 𝗘𝗮𝗿𝗹𝘆 𝗦𝗶𝗴𝗻𝗮𝗹𝘀 𝗶𝗻 𝟮𝟬𝟮𝟰
The idea of 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗔𝗜 started surfacing more clearly in 2024.
It is moving beyond upgrade to chatbots, but as a different way of organising work.
We are looking at AI systems that could observe a situation, translate intent into steps, act across tools, check outcomes, and adjust along the way.
Like many, I found that interesting, but conceptual. I believe most organisations were still interacting with AI through chat interfaces.
AI answered questions while humans remained responsible for execution.
𝟮𝟬𝟮𝟱: 𝗪𝗵𝗲𝗻 𝗔𝗜 𝗘𝗻𝘁𝗲𝗿𝗲𝗱 𝘁𝗵𝗲 𝗪𝗼𝗿𝗸𝗳𝗹𝗼𝘄
In 2025, AI began showing up inside workflows
Organisations started moving beyond copilots and assistants to experiment with AI agents embedded directly into operations.
While chat-based AI responds to questions, agentic systems work toward outcomes. An agent can take a goal, break it into steps, pull data across systems, act within defined boundaries, and validate results as it goes. This allows work to progress continuously, with decisions moving forward without waiting for constant human prompts.
Two developments converged to drive this shift in 2025 :
• 𝗜𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝗰𝗲 𝗕𝗲𝗰𝗮𝗺𝗲 𝗖𝗵𝗲𝗮𝗽𝗲𝗿
According to the Stanford AI Index, the cost of running AI inference at GPT-3.5–level performance has dropped by more than 280× in under two years. This dramatic decline made continuous machine reasoning viable across everyday business activities rather than confined to isolated pilots or innovation labs.
• 𝗔𝗜 𝗕𝗲𝗰𝗮𝗺𝗲 𝗠𝗼𝗱𝘂𝗹𝗮𝗿
As companies move away from single, general-purpose models toward more specialised systems, we are beginning to see the rise of domain-specific agents—finance agents, inventory agents, logistics agents, and beyond. This shift signals a broader architectural change in how work is designed and executed.
𝗪𝗵𝗮𝘁 𝗖𝗵𝗮𝗻𝗴𝗲𝗱 𝗜𝗻𝘀𝗶𝗱𝗲 𝗢𝗿𝗴𝗮𝗻𝗶𝘀𝗮𝘁𝗶𝗼𝗻𝘀
As AI agents moved into day-to-day operations, decisions started happening faster.
In functions like supply chain and inventory management, companies could spot issues earlier and respond without waiting for multiple handovers or approvals. As such, efficiency was no longer a differentiator — it became the baseline.
What set companies apart was how quickly they could react when something changed.
𝗛𝗼𝘄 𝗧𝗵𝗶𝘀 𝗜𝗺𝗽𝗮𝗰𝘁𝘀 𝗩𝗮𝗹𝘂𝗲 𝗖𝗵𝗮𝗶𝗻
As AI dramatically reduced the cost of data collection, reconciliation, reporting, and routine analysis, it squeezed the middle of the value chain.
Many service-based businesses discovered that a large part of their fees had been justified by manual coordination and human effort, not by unique insight or strategic judgment. Once those activities became faster and cheaper through AI, clients started questioning what they were really paying for.
Margins tightened as a result — and that pressure has not gone away.
𝗥𝗼𝗹𝗲𝘀 𝗦𝗵𝗶𝗳𝘁𝗲𝗱, 𝗪𝗶𝘁𝗵𝗼𝘂𝘁 𝗔𝗻𝗻𝗼𝘂𝗻𝗰𝗲𝗺𝗲𝗻𝘁
As more execution work became automated, human roles quietly shifted upstream.
Value no longer came from doing the work, but from deciding what the work should do and where it should stop. The people creating the most impact were those who could set clear intent, govern automated systems, and intervene early when things drifted off course.
This shift was already visible in operations.
In finance, AI agents moved beyond simple detection into investigation, correlating signals across systems and resolving cases with minimal human involvement. Scam detections and escalations are perfect demonstrations.
In supply chains, agentic systems proved their worth during global disruptions: when shipping routes were blocked at the Suez Canal, agents detected the event, recalculated alternatives, assessed trade-offs, and surfaced viable options before teams had fully absorbed the situation.
In this new reality, judgment became the real differentiator.
𝗘𝗺𝗲𝗿𝗴𝗲𝗻𝗰𝗲 𝗼𝗳 𝗡𝗲𝘄 𝗖𝗼𝗺𝗽𝗲𝘁𝗶𝘁𝗶𝘃𝗲 𝗠𝗼𝗮𝘁𝘀
As AI intelligence became widely accessible, traditional advantages weakened.
Simply having advanced software was no longer enough. What began to matter more was whether an organisation could provide AI with its unique context — the accumulated knowledge of how its business really works, the lessons learned from past decisions, and the realities that don’t appear in formal reports.
Equally important was how deeply AI was embedded into core workflows, where decisions are made and actions occur.
Without context, memory, and integration, intelligence remained generic. Software alone stopped being defensible.
𝗙𝗿𝗼𝗺 𝗣𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝘃𝗶𝘁𝘆 𝘁𝗼 𝗟𝗲𝘃𝗲𝗿𝗮𝗴𝗲
By 2025, AI adoption was widespread, yet only a small fraction of organisations described their deployments as mature or fully scaled.
Many had already built or purchased agents across marketing, planning, research, and operations, but struggled to turn these capabilities into lasting advantage.
As intelligence became more accessible, differentiation shifted away from tools and toward how deeply systems were embedded into core workflows, how well they reflected institutional knowledge, and how confidently leaders could govern their use.
Now, the real challenge is about shaping coherence, direction, and intent across the organisation.
𝟮𝟬𝟮𝟲 𝗧𝗿𝗮𝗷𝗲𝗰𝘁𝗼𝗿𝗶𝗲𝘀
As organisations look toward 2026, a few strategic questions need to move onto the leadership radar:
• 𝗪𝗵𝗲𝗿𝗲 𝗶𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝗰𝗲 𝘀𝗵𝗼𝘂𝗹𝗱 𝘀𝗶𝘁 𝗶𝗻 𝘁𝗵𝗲 𝘃𝗮𝗹𝘂𝗲 𝗰𝗵𝗮𝗶𝗻 – Which decisions, handoffs, or coordination points could be restructured if reasoning happens continuously — not episodically?
• 𝗛𝗼𝘄 𝗮𝗴𝗲𝗻𝘁𝗶𝗰 𝗰𝗮𝗽𝗮𝗯𝗶𝗹𝗶𝘁𝗶𝗲𝘀 𝗿𝗲𝘀𝗵𝗮𝗽𝗲 𝘁𝗵𝗲 𝗯𝘂𝘀𝗶𝗻𝗲𝘀𝘀 𝗺𝗼𝗱𝗲𝗹 – What becomes cheaper, faster, or unnecessary — and where new forms of value, leverage, or differentiation emerge as a result?
• 𝗪𝗵𝗲𝘁𝗵𝗲𝗿 𝗰𝗼𝗺𝗽𝗲𝘁𝗶𝘁𝗶𝘃𝗲 𝗮𝗱𝘃𝗮𝗻𝘁𝗮𝗴𝗲 𝘀𝗵𝗶𝗳𝘁𝘀 𝗳𝗿𝗼𝗺 𝗲𝘅𝗲𝗰𝘂𝘁𝗶𝗼𝗻 𝘁𝗼 𝗼𝗿𝗰𝗵𝗲𝘀𝘁𝗿𝗮𝘁𝗶𝗼𝗻 – In a world where many can deploy similar tools, advantage may come from how systems are governed, sequenced, and aligned.
• 𝗪𝗵𝗮𝘁 𝗼𝗿𝗴𝗮𝗻𝗶𝘀𝗮𝘁𝗶𝗼𝗻𝗮𝗹 𝗰𝗮𝗽𝗮𝗯𝗶𝗹𝗶𝘁𝗶𝗲𝘀 𝗻𝗲𝗲𝗱 𝘁𝗼 𝗲𝘃𝗼𝗹𝘃𝗲 – Not just technical skills, but leadership judgment, decision rights, incentive design, and the ability to intervene when automated systems drift.
These are no longer technology decisions, but strategic ones.
(Last published – Dec 2025, by Christina Lim)
Recent Comments