Most AI content gives you a framework. Tiwa York gives you a verdict.
The founder who built Kaidee to 35 million users and guided it to a successful exit sat down with SEA of Startups and said what most operators are afraid to say out loud: your team is probably performing AI adoption, not doing it. And the longer you stay there, the harder it gets to move.
Here’s what he actually said — the numbers, the examples, the provocations.
The 5 Levels of AI Maturity (And Why 1.5 Is a Trap)
Tiwa’s framework runs from 0 to 4. Most conversations stop at listing the levels. The more important conversation is why so many companies get stuck halfway through Level 1.
Level 0 — Unaware: No AI tools in use. Working like it’s 2019.
Level 1 — Curious: ChatGPT is bookmarked. It gets used for emails and translation. Actual work output: unchanged.
Level 1.5 — The Trap: This is where Tiwa spends most of his time on stage. A few people are experimenting. Strategy decks mention AI. But workflows, decisions, and output haven’t moved. He calls this adoption theater — and it’s where the majority of SEA companies currently sit.
Level 2 — Active: AI is genuinely built into daily work. Measurable productivity gains of 25–50%.
Level 3 — Integrated: Multiple AI tools connected in smooth workflows. The data analyst goes from one report a week to one a day. The PM tests ideas overnight with simulated customers. 2–3x productivity — and completely redesigned ways of working.
Level 4 — Transformative: Creating value streams that simply didn’t exist before. Tiwa estimates this is roughly 2% of the global workforce today.
The goal isn’t to inch from 1.5 to 2. It’s to move from 1.5 to 3, and then to 4. Anything less is rearranging deck chairs.
The Mental Model That Changes Everything
Tiwa’s most useful reframe isn’t a framework — it’s a metaphor.
Think of AI as the most capable but most forgetful intern you’ve ever hired. It can do almost anything better than any employee on your team. But the moment it leaves a conversation, it remembers nothing. Zero context. Starting from scratch.
This metaphor matters because it tells you exactly what your job is: you’re not a user of AI. You’re a systems designer for AI. Your task is building the handoff infrastructure — the context-carrying mechanisms, the memory systems, the structured prompts — that prevent that amnesia from killing your output quality.
Tiwa draws a direct parallel to the Toyota Production System. You’re not optimising one conversation. You’re building a manufacturing process for intelligence, with daily standups, continuous improvement loops, and institutional memory that compounds over time.
Most companies treat AI like a vending machine. High performers treat it like a factory floor.
The Numbers That Should Stop You Mid-Sentence
If you think the efficiency gap between good and great AI usage is somewhere between 20–30%, Tiwa has a number for you.
The difference between a 30% productivity gain and a 300x productivity gain isn’t the model you’re using. It’s how you’re using it.
That’s not a typo. 300x. The delta between someone using AI as a faster search engine and someone who has built genuine fluency — with context management, iteration discipline, and system-level thinking — is not incremental. It’s categorical.
On token economics specifically, Kevin cited Jensen Huang’s framing directly: a developer earning $500K annually should be spending roughly $250K a year in AI tokens. That’s the ratio of a high-performance AI-native engineer. For context: serious power users are already spending $500+/month on tokens. Some AI-native startups are at $1,000 per person per day.
If your developers aren’t asking for AI budget, Tiwa’s take is unambiguous: that’s a performance issue.
The Hiring Freeze Argument (And Why It’s Not Crazy)
The most provocative position Tiwa took in the recording:
Freeze all hiring until your AI implementation is complete.
The reasoning is mathematical. Communication pathways explode non-linearly with headcount:
5 people → 10 pathways
10 people → 45 pathways
20 people → 190 pathways
Every person you add before you’ve stabilised your AI workflows creates coordination overhead that compounds. You’re layering human complexity on top of unresolved process complexity. The problems don’t add — they multiply.
The implication for most early-stage SEA founders: your instinct to hire for growth may be the thing slowing your growth. A team of 6 people who are genuinely at Level 3 will outrun a team of 15 people stuck at Level 1.5, every time.
The Middleware Trap: A Warning for Builders
Tiwa is an investor. He’s pattern-matching on where value will be captured — and where it will evaporate.
His verdict on horizontal and middleware AI companies: 18-month obsolescence risk. The major frontier models are absorbing middleware functionality as a matter of course. If your moat is sitting between the model and the enterprise, that’s a shrinking gap.
The defensible positions he sees in SEA:
Vertical solutions with deep workflow integration and hard-to-replicate domain understanding
Regulated, complex legacy environments where switching costs are real and proprietary data is locked in
Physical AI — Tiwa cited MUI Robotics, which has deployed an AI tongue (taste and smell sensors) across dairy companies, water utilities, and hotel renovation monitoring, and is currently running a research project on early liver cancer detection through smell. 300+ clients. 50+ multinationals. That’s not a middleware play.
The common thread: proprietary data, physical integration, or regulatory complexity. If you can be replaced by a model update, you’re not building a business — you’re building a feature.
Two Real Examples, Not Hypothetical Ones
The Jira/Confluence Replacement: A software development house replaced its entire project management stack — Jira, Confluence, the lot — in four days using AI-assisted development. Annual savings: $24,000. More importantly, they own the system now. No vendor dependency. No per-seat pricing. No waiting for a roadmap that doesn’t match their workflow.
The HubSpot Replacement: A friend of Tiwa’s replaced their entire HubSpot instance with a custom-built CRM in eight hours of AI-assisted coding. Eight hours. The off-the-shelf tool cost thousands annually and didn’t fit the workflow. The custom solution does — and it cost a weekend.
The pattern here isn’t “build vs. buy.” It’s “stop buying things that make you dependent when you could own the thing in a day.”
What AI-First Actually Requires From Leadership
Tiwa’s framework for leaders isn’t about tool selection. It’s about accountability architecture.
The key shifts:
Every function owns its own transformation. This can’t live with the CTO alone. Engineering, product, marketing, finance, customer success — every team lead is responsible for their own AI integration roadmap.
Model the behaviour publicly. If leadership isn’t visibly using AI — and visibly failing with it, learning from it, sharing what they found — no one else will take the cultural signal seriously.
Measure outcomes, not activity. Logins aren’t fluency. Licenses aren’t execution. The metrics that matter: workflow velocity, decision speed, output quality. Not hours of AI training completed.
Daily continuous improvement. Not a quarterly AI review. A daily standup cadence for what’s working, what broke, what gets refined tomorrow. Toyota didn’t build the production system in a sprint. Neither will you.
The Real Question
Tiwa closed with the line that stayed with everyone in the room.
“The question isn’t how do we find extraordinary people. It’s whether extraordinary people get unleashed inside this org — or leave to do it on their own.”
For founders in SEA: you probably already have the talent. The judgment is in the building. The only variable is whether you build the systems that let it operate at full power — or whether you stay at Level 1.5 long enough that the people who figured it out first come back to compete with you.
Watch the full conversation with Tiwa York on SEA of Startups
