Setting the Stage

Every boardroom is buzzing about AI. According to McKinsey, nearly 80% of organizations are engaging with AI — 35% have systems in production, another 42% are piloting. Gartner reports that over 60% of boards believe AI will be the single greatest driver of shareholder value within the next two years.

But beneath the enthusiasm lies a sobering truth: 95% of GenAI pilots fail to scale. 30% of projects are abandoned entirely. Internal AI teams often cost over $1 million a year, yet still fail to deliver outcomes. What’s missing isn’t ambition — it’s the shift from generic tools to Agentic AI, powered by embedded expertise.

That paradox — massive potential, paired with frustratingly low success rates — framed the live conversation I had with our Chief AI Officer, Anirudh Badam, and our Forward-Deployed Engineer, Vijay Sagar.

What follows isn’t a replay of slides. It’s a distillation of what we learned in the discussion: the real reasons enterprises are stuck, and the practical steps leaders can take to move from endless pilots to production outcomes.

1. Adoption ≠ Outcomes

“Adoption doesn’t equal impact,” Anirudh cautioned early. “You can have pilots everywhere and still fail to get value in production.”

That one line set the tone. Enterprises are awash in pilots, but only a fraction produce measurable results. Finance and tech have surged ahead, largely because they can recruit and retain top AI engineers. Healthcare, insurance, logistics, and other non-tech industries are playing catch-up without the same bench strength.

‍

Action for leaders: Audit your own AI portfolio. Count the number of pilots you’ve launched in the past 18 months. Now ask: Which are delivering measurable business value today? Which are still “demo-ware”? That simple exercise often reveals how wide the gap is between enthusiasm and execution.

2. Beyond the Myth of the “Talent Shortage”

Boards tend to describe the problem as a talent shortage. “We need more AI engineers.” “There aren’t enough skilled people.”

But as Anirudh challenged, this framing misses the mark.

“It’s not about headcount. You can hire brilliant PhDs and still fail once you go live. The real gap is production expertise.”

The distinction is subtle but vital. Research labs and academic institutions train people to push the boundaries of AI. But deploying those models into the messy realities of enterprise systems — compliance rules, brittle integrations, shifting data quality — that requires a completely different skillset.

Vijay sees this daily as a Forward-Deployed Engineer:

“Most AI talent is commoditized now. The difference comes from applied experience — taking proven patterns across industries and embedding them where it matters.”

This is why so many organizations believe they’ve solved for talent when, in fact, they’ve only staffed for experimentation.

Action for leaders: Stop asking “Do we have AI talent?” and start asking, “Do we have AI production experience?” The answers may be very different.

3. Why Traditional Approaches Collapse Under Pressure

For years, outsourcing was the go-to solution for technology capability gaps. Need a mobile app? Hire a dev shop. Need a new CRM integration? Bring in a systems integrator.

But GenAI is a different animal. The old playbook breaks down.

“Consulting firms can write a strategy doc. IT can wire up an API,” Anirudh said. “But that doesn’t mean you have a production-grade agent that scales.”

What looks like progress often isn’t. Vijay recounted:

“I’ve watched teams spend months on a demo that looks great — until you ask about drift monitoring. That’s when projects collapse.”

The reason is simple: pilots live in controlled sandboxes. Production lives in chaos. If you don’t plan for the chaos — governance, monitoring, retraining — the first spike in usage or drift in data topples the project.

Action for leaders: Pressure-test your partners. Don’t just ask for a demo. Ask them how they’ll handle regression testing, monitoring for drift, and compliance guardrails. If they can’t answer, they’re not ready to carry you across the production chasm.

Want a deeper framework?
We recently published The Top 7 Criteria for Evaluating Agent Builder Platforms — a practical guide to pressure-testing vendors and ensuring they’re ready for production.

4. Rethinking ROI in AI

Every leader wants to know: what’s the return?

The paradox is that while AI promises trillions in economic value, the economics of AI projects inside most enterprises are brutal. Internal AI teams are expensive — often costing $1M+ per year once you add up salaries, cloud usage, security reviews, compliance overhead, and the opportunity cost of delay. And despite the spend, Gartner data shows that 30% of GenAI projects are abandoned before they ever reach production. MIT puts the broader failure rate at 95%.

That’s why the traditional ROI math doesn’t hold. If you measure ROI only as “(Return – Cost) ÷ Cost,” you miss the two variables that make or break AI value: speed and reliability — the hallmarks of Agentic AI done right.

As Anirudh explained in the session:

“ROI in AI comes down to three things: time to production, success rate, and cost. Internal builds are slow, expensive, and unreliable. Forward-Deployed Engineers flip that equation.”

A Worked Example: Internal Team vs. FDE Model

Scenario 1 — Internal AI Team

• Cost: 100% baseline (salaries, infra, governance, compliance, delays)
• Time to Production: ~12 months
• Success Rate: ~50%
• Risk: High sunk costs if project stalls or fails
  
  

Scenario 2 — FDE Model

• Cost: ~50–60% of baseline (leaner, more focused engagement)
• Time to Production: 3–4 months (≈70% faster)
• Success Rate: 80%+ (applied expertise, proven patterns)
• Risk: Lower — faster path to ROI, less wasted investment

Effective ROI: Faster deployment, higher likelihood of success, and lower total spend. Even if the cost is half a million, the speed to outcomes means the enterprise starts realizing value 8–9 months sooner than the internal path.

Action for leaders: Run this comparison for your own organization. In Column A, capture the full cost of internal builds, including the risk of failure. In Column B, map the cost and outcomes of an embedded model. When you see the numbers side by side, it’s clear: the faster you get to production with higher odds of success, the higher your real ROI.

5. From Pilots to Infrastructure

The most powerful moment came at the close.

“Don’t get stuck in endless pilots,” Ani said. “Treat AI adoption like building core infrastructure — because it is. The companies that win will be the ones who invest in the right expertise, not just the right tools.”

This was more than advice. It was a challenge. Pilots are safe. Infrastructure is commitment. But that’s the leap every enterprise will have to make if they want to capture their share of the trillions in value AI is set to unlock.

Action for leaders: Pick one initiative. Stop treating it as an experiment. Wrap it with governance, performance KPIs, and adoption goals — and run it like infrastructure.

Final Thoughts

The lesson from this session is clear: AI’s winners won’t be the companies that experiment the most. They’ll be the ones who scale the fastest, with production expertise embedded from day one — and who embrace Agentic AI as core infrastructure, not just another tool.

If you’re a leader reading this, here’s your action list:

1. Audit your pilots for outcomes.
2. Look at your talent — do you have production expertise, or just résumés?
3. Pressure-test vendors on governance, drift, and monitoring.
4. Run the ROI exercise — internal vs embedded.
5. Stop piloting. Start building infrastructure.

Because in AI, every quarter you wait isn’t neutral. It’s lost ground.

Watch the full session on-demand
This blog distilled the highlights, but you can catch the entire conversation with Garrett, Anirudh, and Vijay — including the live role play demo — here:
Unlocking AI for Every Enterprise: The Power of Forward-Deployed Engineers

‍