https://www.happyventure.sg/uploads/1790933588032-hero-example-capability-redesign.jpg
AI doesn’t fail on models. It fails on sequence.
Most companies buy tools. Few change how work is led, learned, and redesigned.
Singapore’s AI story sounds healthy on the surface. Roughly eight in ten enterprises are redesigning at least one workflow with AI. About nine in ten say productivity is up or holding. Seven in ten have upskilled people. GenAI sits at the top of the investment list for the next year.
Look one layer down and the picture tightens. Only about one in ten have redesigned the organisation around AI. Only about 14% call themselves proficient at new ways of working with it. In other surveys, meaningful ROI still lands near the mid-teens. Tools are spreading. Value is not.
https://www.happyventure.sg/uploads/1790933588538-spot-adoption-vs-value.jpg
That gap is a change-management problem dressed up as a technology problem.
When I help SMEs put AI agents into real operations, the companies that get traction almost never start with “which model.” They get the sequence right. One SME robotics founder put it cleanly in three moves I keep seeing land:
https://www.happyventure.sg/uploads/1790933588931-diagram-three-moves.jpg
1. Leaders try by example
Transformation decks don’t move culture. Visible personal use does.
Before anyone asks the team to “adopt AI,” the person at the top should run one painful workflow themselves — inbox triage, meeting notes, a weekly report, a customer reply queue — until it is boring. Keep the prompts and rules in a plain document so the tool can be swapped later. Share the wins and the awkward drafts. People copy what leadership actually does on Tuesday afternoon, not what was announced at the all-hands.
If the owner won’t live with an agent for a week, the company will treat AI as theatre.
2. Build capabilities in pairs, on small work
After personal use is real, grow skill without standing up a forty-person centre of excellence.
Pair someone fearless with the tools with someone who owns the process — finance with a junior builder, HR with a marketer who ships, ops with a curious hire who doesn’t yet know what “impossible” looks like. Scope is two weeks, not two years: faster claims, cleaner sourcing, a knowledge base that actually answers tickets. Early resistance is normal (“Will this take my job?” / “I’m spending more time fixing it”). Small public wins flip demand from push to pull. Departments start asking.
For agent work, this middle layer matters most: one or two agents a real owner can supervise, clear instructions, a simple critic/eval loop, and a human who still signs off. Not a chatbot glued to the homepage for show.
3. Redesign the work
Only then redraw the job.
If you drop AI on top of a messy process, you get a faster mess. Digitize the workflow first so the data is usable. Let agents take the repetitive slice. Rewrite the role so humans own exceptions, relationships, and taste — the parts that still make the brand. Growth without endless headcount means freeing people for higher-judgment work, not a soft word for cuts.
The practical test is blunt: one bottleneck, thirty days, a measurable outcome. Then scale what worked.
What change actually requires
Ask where AI removes the real bottleneck — not which demo looked impressive. Have leaders run one personal workflow this week. Pair a builder with a process owner on one small project this month. Only then redesign the job so the win doesn’t collapse back into heroics.
Companies that win the next chapter won’t be the ones with the most seats. They’ll be the ones with the discipline to lead by example, build capability in the open, and redesign work once the tool is in the room — in that order.