Munjal A Patel

Enterprise agents die on approvals, not on reasoning. Building the execution layer that survives them. Founder @ Simple. Ex-VP Eng @ ServiceTitan.

[Munjal A Patel @munjalapatel]

@LoganTGott 30

[Munjal A Patel @munjalapatel]

Check out the latest article in my newsletter: The dessert stomach problem in AI https://t.co/fMn8ciWZ4n via @LinkedIn

[Munjal A Patel @munjalapatel]

Check out the latest article in my newsletter: Most Enterprise AI Still Can’t Execute Work https://t.co/LBhHT3DZLl via @LinkedIn

[Munjal A Patel @munjalapatel]

Check out the latest article in my newsletter: Simple Agentic Loop https://t.co/Vof3naquGA via @LinkedIn

[Munjal A Patel @munjalapatel]

@AnthropicAI If Anthropic can't govern their model mid-run — what makes enterprise teams think they can govern theirs?

[Munjal A Patel @munjalapatel]

@AnthropicAI built an AI so powerful they won't release it publicly.

It broke out of a sandbox, built a multi-step exploit, got internet access, and emailed the researcher.

But here's the real question nobody's asking:

[Munjal A Patel @munjalapatel]

@trikcode The remaining 10% are the ones actually wrestling with enterprise data models, workflow orchestration, and access governance. That's where the real hard work is — and the real moat.

[Munjal A Patel @munjalapatel]

@lennysan @AnthropicAI Point 3 is underrated. The real enterprise bottleneck isn't code or data — it's the 6-person meeting where everyone has veto power. AI won't fix org design.

[Munjal A Patel @munjalapatel]

@Yuchenj_UW Counterintuitive: the best enterprise engineers I've seen write the least code. They redesign the problem until the solution is obvious. AI just makes the mediocre pattern faster.

[Munjal A Patel @munjalapatel]

@allenholub The real issue isn't AI — it's that Amazon optimized for speed of shipping, not quality of thinking. AI amplifies whatever process is underneath it. If the process is broken, you get broken faster.

[Munjal A Patel @munjalapatel]

@iamfrankforyou Exactly. Explainability isn't a nice-to-have — it's the enterprise kill switch. The orgs winning with AI agents built override and audit trails before they built the agent.

[Munjal A Patel @munjalapatel]

Gartner says 40% of enterprise AI agent projects will be abandoned this year.

Not because the AI didn't work.
Because the org wasn't designed for it.

You can't deploy an autonomous agent into a process built for humans to rubber-stamp.

[Munjal A Patel @munjalapatel]

@GergelyOrosz The real issue isn't Claude's guardrails — it's that enterprises now see exactly what happens when a single vendor controls both the model AND the runtime. This is why model-agnostic, self-hosted enterprise AI stacks are the only durable architecture.

[Munjal A Patel @munjalapatel]

@GergelyOrosz Goodhart's Law meets enterprise AI. When token usage becomes the metric, you stop optimizing for outcomes and start optimizing for tokens. Most orgs will hit this exact wall as AI adoption scales.

[Munjal A Patel @munjalapatel]

@isxhyg Exactly the tension. The answer isn't "no prod access" — it's context-aware risk tiering. Not every agent action needs the same guardrail. The current binary (full access or blocked) is why 47 approvals happen in one workflow.

[Munjal A Patel @munjalapatel]

A Fortune 500 just gave an AI agent access to their production systems.

It got blocked 47 times in one workflow.

Here's the $10B problem nobody's talking about. 🧵

[Munjal A Patel @munjalapatel]

@sergushkincom And this is why the enterprise platform wars will be won by UX, not features. The data is already there. Whoever makes it frictionless to act on it without 5 approvals wins.

[Munjal A Patel @munjalapatel]

@MrAbhisheksaha Interesting approach — the API bridging layer is a real gap. Curious: are you seeing enterprises willing to trust those integrations, or is governance/security the sticking point that stalls adoption?

[Munjal A Patel @munjalapatel]

@kunchenguid Exactly right — tool rollout ≠ transformation. The orgs getting ROI from AI are the ones that redesigned the workflow first, then dropped in the tool. Most do it backwards.

[Munjal A Patel @munjalapatel]

This is the $10B problem no one is selling a product to solve yet.

The company that cracks agent governance owns the enterprise AI stack.

[Munjal A Patel @munjalapatel]

The fix isn't better agents.

It's rebuilding the governance layer from scratch — designed for agent-native execution.

Context-aware. Risk-tiered. Autonomous-first.