Leading AI Teams

Org design kills more AI programs than bad models

Here's a secret nobody in the AI hype cycle wants you to hear: the reason most AI initiatives fail has nothing to do with AI. It's the org chart. It's who reports to whom. It's whether the VP of Product and the VP of Engineering had lunch last Tuesday.

Think about it this way. Imagine you're 10 years old and you want to open a chain of pizza restaurants. You've got the world's greatest pizza recipe (that's your AI model). But if you try to open 50 locations on day one without a central kitchen, without a supply chain, without training manuals — you'll have 50 locations making 50 different pizzas, burning through cash, and probably giving someone food poisoning.

That's exactly what happens with AI teams. Competitors don't pull ahead because they hire better engineers. They pull ahead because they structure their teams to ship.

Your job as an executive isn't to understand transformers or fine-tuning. Your job is to build the restaurant franchise system — then let the chefs cook.

The 4-stage pizza franchise (a.k.a. AI maturity model)

Every company moves through four stages. Think of them like growing a pizza franchise from a single oven in your garage to a national chain.

Notice the arrows. Each one says "Exec decision" — not "hire" or "buy tool." You are the bottleneck and the unlock. Let's break each stage down.

The maturity stages at a glance

There Are No Dumb Questions

Disaster theater: when stages go wrong

Company A: The stage 2 trap (a.k.a. "we built a kitchen but never opened any restaurants")

Company A built a beautiful centralised AI platform. State-of-the-art tooling. Gorgeous dashboards. One problem: they stayed there for two years.

The central team became a bottleneck so massive that product teams waited six months for a single AI feature. The intake queue grew to 47 tickets. Engineers started calling the AI platform team "the Department of No." Meanwhile, three competitors shipped AI-powered features and stole 12% market share.

The CEO's post-mortem said "technology failure." The real diagnosis? An org design decision killed it. They built the central kitchen but refused to open franchise locations. The chefs sat in HQ perfecting recipes that never reached a customer.

Company B: The stage skipper (a.k.a. "we opened 50 restaurants with no recipe book")

Company B looked at Company A's slow pace and said "not us." They jumped from Stage 1 straight to Stage 3 — embedding AI engineers into product teams without building a platform first.

The results were spectacular. Spectacularly bad.

• 14 different LLM providers in production (fourteen!)
• 3 PII incidents — customer data hitting unvetted third-party APIs
• No shared cost dashboard — the CFO discovered $180k/month in AI API spend buried across 23 different credit cards
• Zero reusable components — every team built their own prompt templates, their own logging, their own everything

The board found out about the PII incidents from a journalist. Speed without structure doesn't land on your roadmap — it lands on the board agenda as a liability.

Think about it: 50 pizza restaurants, each buying their own flour from whoever, no health inspections, no recipe standards. One of them serves raw chicken. Now it's on the evening news and all 50 locations have a problem.

The hero's journey: Priya's 90-day plan

Priya Anand, VP of Engineering at a 400-person fintech, watched Company A and Company B crash in real time. She had both their post-mortems pinned to her office wall.

Her situation: Stage 2, with a ticking clock. 90 days until a major contract renewal that required demonstrable AI capabilities in the product. Lose the contract, lose 30% of revenue. No pressure.

Month 1: The audit that changed everything (Days 1–30)

Priya started with a question that made her engineering leads uncomfortable: "Show me every AI API call we make and every dollar we spend on it."

The answer was worse than she expected:

• 6 different LLM providers already in production (she'd only approved 2)
• $47,000/month in AI API spend scattered across individual engineering budgets — invisible to finance
• Prompt templates saved in 4 different Notion workspaces, Slack threads, and one engineer's personal GitHub repo

The exec decision: One approved provider. One dashboard. Hard cutover deadline — 30 days, no exceptions.

Result after Month 1: $47k/month dropped to $31k/month. Not because they cut capability — because they eliminated duplicate providers, negotiated volume pricing, and stopped paying for 6 different rate limits.

Months 2–3: The embedding sprint (Days 31–90)

With the platform locked down, Priya made her second exec decision: embed AI engineers into the top 3 product teams. Not all teams — just the three closest to customer-facing features for the contract renewal.

She built a shared API wrapper — a single internal connection point that all teams use to call the AI service — with logging baked in. Every API call, every token, every cost automatically tracked. Compliance checks ran on every request. No team had to think about it.

The turning point: Week 6. The first embedded engineer shipped a customer-facing feature in 4 days. The centralised team had estimated it at 8 weeks.

4 days vs. 8 weeks. Same engineer. Same skills. Different org structure.

That single data point gave Priya the internal proof to accelerate the remaining embeddings. Within 6 months: feature velocity ran 3x faster, and the fintech won the contract renewal.

The Priya playbook: what to steal

The pattern that always holds

Think back to pizza. The central kitchen (Stage 2) creates the recipes, the supply chain, the quality standards. The franchise locations (Stage 3) serve the customers. You need both. In that order.

Try it

Back to Priya

The 90-day clock ran out. Priya walked into the contract renewal meeting with a working demo — a customer-facing AI feature that an embedded engineer had shipped in four days.

The client's procurement lead leaned forward. "Your competitors said they'd need eight weeks to build something like this. You did it in four days?"

"Same engineer," Priya said. "Different org structure."

They signed. Priya kept the post-mortems of Company A and Company B pinned to her wall.

Key takeaways

• Build the kitchen before you franchise. Use the centralised AI team to build the shared platform — but once it's built, push AI engineers into every product team. Don't keep them central.
• Skipping stages is not speed — it's debt. Every time you embed AI engineers before the platform layer is ready, you create fragmentation, cost chaos, and compliance gaps. Just ask Company B.
• You are the unlock. Each maturity stage advances with a single exec decision — you don't need to wait for a new hire, a new tool, or a new budget cycle. The decision comes first; everything else follows.