This deep dive expands on the Loop Asia conversation with Manjunath Bhat, Distinguished VP Analyst, Gartner.

Somebody told Manjunath Bhat that you're not using AI to its fullest potential unless you're spending a thousand dollars per engineer. He can't place the source, but he doesn't dispute the direction: agentic coding is expensive, the cost curve rises with every model release, and most executives approving these budgets don't yet have a framework for governing the spend. That gap — mandate to deploy, no governance to control it — is the terrain a COO, Head of Professional Services, or CEO needs to own before the bill arrives.

Three mismatches, three decisions

The cadence mismatch is productivity outrunning security: engineering ships faster than governance was built to review. Bhat's fix isn't slowing engineering down — it's platform engineering absorbing the gap, building guardrails that let speed and security move at the same pace.

The impedance mismatch is more fundamental. The applications your business depends on are expected to behave deterministically — same input, same output — but the tools building large portions of that code are non-deterministic. No amount of prompting closes that gap; it has to be managed architecturally, through verification and validation.

The accountability mismatch should worry a Head of Professional Services most directly. Engineers, and the leaders who sign off on their work, remain accountable for artifacts an agent produced that they may not have reviewed in full. "We are expected to continue to take accountability even if we have not done the work," Bhat said. That's an organisational design problem, not a technical one — who signs off, on what basis, with what visibility into what the agent did.

Bhat's structural answer to all three is the same: platform engineering becomes the translation layer. Finance sets the guardrails — cost ceilings, usage policy. Platform engineering converts that into something engineers consume self-service, so autonomy and control stop fighting each other line by line.

Don't repeat the cloud computing mistake — but not for the reason you'd expect

I asked Bhat whether agentic coding is heading toward the same trap as early cloud: everyone assumed "pay for what you use" would mean lower costs, and instead usage — and spend — went up. He agreed with the premise, then reframed the conclusion. Nobody who used cloud well actually cut their bill; the ones who got value used the elasticity to build things that weren't previously possible. He expects the same curve here: costs keep rising release over release, and organisations chasing cost reduction alone will find the economics don't hold up. The ones building products that were previously commercially unviable are the ones who can justify the spend — which makes "how do we control AI coding spend" the wrong question in isolation. The real pairing: what capability does this spend unlock, and does the governance exist to control it while you chase that capability.

Where this lands for leadership

Bhat's three mismatches double as a diagnostic for a mandate Blue Connector sees often: board-level pressure to deploy AI, budget behind it, no framework yet for governing the spend, the accountability, or the architecture underneath. If you can't currently name who's accountable when an agent's output ships, whether platform engineering has authority to set cost guardrails, or whether validation capacity is keeping pace with generation capacity — that's the gap to close before scaling spend further, not after.