When we hosted our webinar The New AI Reality: Stop Building Everything and Start Building Smart, the goal wasn’t to dazzle with demos. It was to cut through the noise.

The AI market today is full of shiny objects—promises that with the right model or a handful of engineers, any enterprise can “agentify” overnight. But the truth is more complicated. Most pilots stall. Most budgets balloon. And most leadership teams eventually ask the same hard question: why are we spending so much time and money without seeing results?

That’s where this conversation started. Together with Anirudh Badam, our Chief AI Officer, and Gajanan Sabhahit, our Head of Product, we broke down the messy reality of AI adoption and offered a practical framework for companies who want to scale without wasting years or millions.

Why “Build vs. Buy” is a trap

For decades, enterprise leaders have faced the same decision: build in-house or buy off the shelf. At first glance, it makes sense. Build gives you control; buy gives you speed.

But as Anirudh pointed out, AI breaks that binary.

“Neither extreme works. If you buy everything, you get stuck with rigid tools. If you build everything, you’re signing up to maintain an entire stack that has nothing to do with your core business.”

Think back to the early 2000s. Companies poured resources into building custom CRMs and ERPs, convinced they could do it better themselves. Eventually they realized they were reinventing the same plumbing—and SaaS exploded to solve the problem.

‍

We’re at the same point with AI. The question isn’t “build or buy?” It’s what platform do you build with?

‍

The iceberg of AI adoption

Most executives only see the tip of the iceberg: the slick demo where a chatbot answers a question or an agent automates a workflow. That’s the visible 10%.

But beneath the surface is the hidden 90%: data pipelines, observability, governance, API maintenance, evals, memory, compliance.

“Most people think about the model. But 90% of the work is plumbing—data prep, guardrails, observability, memory, API wiring. Only 10% is what actually makes your experience different.” — Gajanan Sabhahit

And here’s the kicker: it’s that hidden 90% that determines whether your project scales—or stalls. Demos don’t show what happens when an API changes, when a prompt fails, or when a regulator asks for audit logs. But those are the realities that decide whether an AI initiative ever makes it to production.

Why DIY collapses under its own weight

‍

So why do so many “do it yourself” approach

es fail? Because what looks simple in a whiteboard session—hire a few engineers, wire up APIs—quickly becomes a treadmill.

• Headcount: To do it right, you don’t just need a couple of engineers. You need a bench of 7–10 specialized roles: LLM engineers, data engineers, infra experts, UI/UX developers, prompt specialists.
• Maintenance treadmill: APIs break, frameworks update, models change. Entire quarters can vanish just firefighting dependencies.
• UX gap: Enterprises can’t ship walls of text. They need streaming, contextual, branded experiences. That requires design expertise most teams don’t have.

And the economics are brutal. Building enterprise-grade agents in-house typically runs $1–$5 million upfront, with another 20–30% annually in maintenance.

As Gajanan put it bluntly:

“DIY sounds simple. In reality, it’s a treadmill of maintenance.”

The cost cliff: what it really takes to build a team

To make this tangible, we walked through the real cost of building an AI team. Just five essential roles—observability engineer, conversation specialist, UI/UX developer, core LLM engineer, data engineer—will run close to $900k a year. Add overhead and benefits, and you’re looking at $1.5M+ annually.

“The cost cliff is real. Most teams don’t see it until they’re already halfway off the edge.” — Garrett Moedl

This isn’t to say enterprises shouldn’t invest in AI talent. But it’s a warning: without a clear strategy, you end up pouring millions into solving problems every other company is already solving.

Lessons learned: what good looks like

So what does success look like? Not in theory, but in practice.

When we look across companies that are succeeding with AI, some common lessons emerge:

• Start lean: The strongest teams don’t try to staff entire AI departments. They empower one or two world-class engineers who focus on what truly differentiates their business.
• Leverage platforms: Instead of building memory, observability, or API connectors from scratch, they use platforms that abstract away the heavy plumbing.
• Move fast to production: Demos don’t create value. Production deployments do. The best teams prove value in weeks, not quarters.
• Unify the experience: Rather than stitching together multiple disconnected bots, they consolidate into a single framework that users can trust.
• Measure outcomes, not models: Success isn’t about whether the model works—it’s whether the business sees impact.

These aren’t just best practices. They’re survival tactics. They show what “good” adoption looks like in an environment where speed, compliance, and scale all matter.

Want the full discussion?
This recap only scratches the surface. To hear the complete conversation with Anirudh Badam and Gajanan Sabhahit, including audience Q&A and live customer examples, watch the full webinar replay here: Watch the Webinar

Smarter economics: the 80/20 model

That brings us to a smarter model for AI adoption. Instead of spreading budgets across 8–10 hires, put:

• 80% of resources into one or two exceptional engineers who deeply understand how to apply AI in your business.
• 20% of resources into a platform that handles the 90% plumbing—guardrails, memory, observability, connectors, governance.

“Don’t spread budget across 8–10 hires. Put 80% into one or two world-class engineers, and 20% into a platform that takes care of the 90% plumbing.” — Anirudh Badam

It’s not about replacing your team—it’s about letting them focus where they add the most value.

Final thoughts

The story of AI adoption isn’t about technology alone. It’s about economics, prioritization, and where companies choose to spend their limited energy.

Here’s what the session made clear:

• Build vs. Buy is obsolete. The future is Build With.
• 90% of the work is hidden. Ignore it, and you’ll stall.
• DIY costs millions and rarely scales.
• Smarter models exist. Lean teams + platforms unlock adoption faster.
• Good adoption looks like production, outcomes, and scale.

That’s the new AI reality. If you’re considering your own path, take these lessons as a guidepost: stop building everything, and start building smart.

👉 Want to see how this could work in your organization? Request a demo.

‍