Once an AI agent is live, it has to thrive in an environment of constant change.

Commentary HR & Education Information Technology Staff Reporter , Singapore Published: 7 minutes ago Photo by snowing via Magnific When Singapore's agentic AI ‘chefs’ arrive, will the kitchen be ready for them? By Zachary Wang Once an AI agent is live, it has to thrive in an environment of constant change. Singapore’s plan to train 40,000 tech professionals in agentic artificial intelligence (AI) by 2029 is a necessary commitment to the future of business.

But having worked through enough enterprise AI deployments, the question that keeps returning is what this new generation of AI specialists will find when they arrive, ready to deploy. In my experience, the answer is rarely a ‘kitchen’ that is ready for them.A Michelin chef cannot perform their best in an empty kitchenMIT Sloan research found that 95% of AI pilots never reach production.

Typically, the reason is not a shortage of skilled engineers. What they find at the enterprise level is an environment that is not structured or prepared for deployment.Common scenarios include disconnected data systems, processes that have not been designed with the precision AI requires, and years of accumulated decision-making or exception-handling that live in the heads of experienced staff, passed on only through informal mentoring.

Without sufficient data and operational procedures, AI systems cannot learn enough to perform. Think about what it would be like to hand a Michelin-trained chef an empty, unequipped kitchen and asking them to run a full dinner service at the same award-winning standard. The chef is not the constraint, the kitchen is.

From observation, this is the situation most enterprises find themselves in when they deploy AI, and the challenge is that many have yet to develop the foundations that will allow agentic AI systems to succeed.Running a kitchen is not the same as being able to cookThere is a second gap that has not received adequate attention. Building a dish brilliantly once is not the same as running a kitchen that produces it consistently.

This requires systems for quality control, processes for managing the people who execute the work, and the discipline to maintain standards during peak periods or when things go wrong.Once an AI agent is live, it has to thrive in an environment of constant change. Systems upgrade, policies shift, products are updated, or customer behaviours may evolve.

An agent that performs well at launch can quietly deteriorate if there is no active management. There must be performance monitoring at scale, constant diagnoses of where the system is producing poor outcomes, as well as updates to the logic and knowledge it draws on. Changes also need to be tested before they reach customers.

This is the responsibility of the head of kitchen operations rather than the chef. A world-class restaurant is built not on culinary talent alone but also the systems that allow culinary talent to be channelled consistently, at volume, over time. In an enterprise AI context, this operational function is nascent at best in most organisations.

Recognising that it needs to exist, and investing in building it is as important as training AI specialists.The restaurants that opened before you are already aheadOperational excellence in a kitchen accumulates through experience. A restaurant running a well-structured kitchen for several years has refined its supply chain, reduced waste, trained its staff, documented its recipes to the gram, and built the required institutional knowledge.

When that Michelin-trained chef walks through the door, the kitchen is ready for them.A restaurant that hires the same graduate into an unstructured environment gets something different: Real talent producing inconsistent results, with no system in place to diagnose why or improve over time.Enterprises that begin building their AI infrastructure now, even in limited domains, are accumulating the equivalent of that institutional knowledge.

Each deployment reveals something about their processes that planning alone cannot teach. Each edge case that surfaces is a lesson about data quality, decision logic, or where the recipe needs to be written more precisely, and each iteration makes the next deployment faster and more reliable.By 2029, organisations that started this work in 2025 and 2026 will have kitchens ready to receive Singapore’s new AI cohort.

Those that wait for the talent to arrive before thinking about the kitchen will be starting from scratch while their competitors are already ahead.Every great kitchen begins with mise en placeSo, what does the preparation actually require?The first task is treating data as an operational asset.

This means auditing existing data, knowing where it lives, whether it is clean and structured, and whether it can be accessed in a form AI can learn from. Most organisations find significant gaps at this stage and finding them early is far preferable to discovering them mid-deployment.The second is process design.

Implicit know