The Data Source #27: The Next AI Infrastructure Layer Will Be Grounded in Code ✨

Welcome to The Data Source, your monthly newsletter covering the top investment themes across cloud-infrastructure, developer tools and data.

Subscribe now and never miss an issue 🦋

---

Over the past two years, AI has swept through enterprise workflows with tools that promise to automate and accelerate everything from sales enablement to customer support. While LLMs have proven effective at deflecting basic questions and generating quick responses, they struggle with the operational depth required for true enterprise-grade use cases.

One of the clearest examples of this gap is in post-sales support, particularly in L2 and technical support environments where frontline automation breaks down. At this level, the cost of resolution is no longer a matter of search or summarization. It’s a question of access: Can a system understand what the product is doing in real time? Can it trace unexpected behavior across a complex, evolving codebase? Can it reason over the source of truth, not just its derivatives?

This is where we believe a new class of AI infrastructure will emerge, one grounded not in documentation or metadata, but in live code and runtime behavior.

Code is Product Truth

Support teams don’t struggle because they lack documentation. They struggle because documentation isn’t the product. It’s a proxy and often an outdated or incomplete one. The actual truth of what the product does, how it behaves, and why a customer is experiencing an issue lives in the codebase, in configuration, in feature flags, and increasingly, across dozens of interdependent services.

This “truth gap” becomes especially acute in high-growth B2B SaaS companies where:

• Product velocity outpaces documentation

• Environment-specific behaviors (e.g., config drift, version mismatches) go undetected

• Engineering resources are scarce and increasingly protected

• Support teams escalate not because they want to, but because they have no viable alternative

And this gap is only widening. With the rise of code-generation tools, features are shipping faster than ever and often without human-authored documentation to match. The more software is written by machines, the more critical it becomes to build systems that can interpret and reason over code directly.

We believe AI agents that can read, reason over, and respond using live system context, including source code will become indispensable infrastructure for enterprise post-sales orgs. Support is just the beginning.

Support Is the Wedge, But Not the Endgame

Support is uniquely painful because poor system understanding becomes a direct business cost:

• Ticket escalations slow down engineering and delay resolution

• Customers churn when trust is eroded by slow or inaccurate responses

• Internal teams waste cycles coordinating instead of solving

But the opportunity runs deeper than ticket resolution. The next layer of AI-native enterprise infrastructure will look more like a shared context engine, one that every post-sales function can tap into:

• Onboarding teams gain visibility into customer-specific feature behavior without pulling in engineering

• Technical Account Managers understand how custom configurations interact with global defaults

• Solutions engineers validate integration questions with grounded, code-aware responses not stale wikis

Over time, this builds toward a foundational layer where real system behavior is queryable, traceable, and shareable across teams, timeframes, and accounts.

Call for Startups

I’m especially excited about teams rethinking how post-sales teams interact with software not through tickets or dashboards, but through what the system is actually doing in real time.

The most compelling companies in this space don’t treat code as something only engineers should touch. They treat it as the authoritative context layer for every customer-facing function. They’re building systems that can ingest messy, sprawling monorepos, trace dependencies across services, and surface answers grounded in how the product behaves today not how it was supposed to behave last quarter.

But technical depth alone isn’t enough. These systems must integrate into real-world workflows embedded directly into Zendesk, Jira, Slack and feel native to the tools teams already rely on. If your product requires users to swivel-chair between dashboards or interpret raw outputs, it’s already out of the loop.

We’ve seen the value of this approach firsthand. A senior technical services leader at a major enterprise software company put it simply: “Each human-resolved ticket costs us about $100. If AI can bring that down to $10, that’s $630,000 in savings a year.” That’s not just a workflow improvement that’s a clear, bottom-line result.

And of course, trust is non-negotiable. Any system operating over production code carries inherent risk. Security can’t be bolted on later. Access models must be scoped, explicit, and auditable from day one. Engineering leaders will scrutinize who touches what, when, and how and if that story isn’t airtight, the deal won’t close.

Most importantly, we’re looking to back founders who see support as a wedge not the ceiling. The same infrastructure that resolves L2 tickets today can power onboarding visibility, customer success insights, and solutions engineering workflows tomorrow. The real opportunity is to build a shared interface for how products behave in production and how every customer-facing team responds.

If you’re building toward that future, I’d love to talk.

The Market is Noisy. That’s Your Opportunity.

The “AI for support” space is crowded with chatbot wrappers and shallow copilots. Most are focused on L1 deflection and lack the technical depth to handle real post-sales complexity. But in that noise lies a clear opportunity: to build systems-aware tooling that becomes mission-critical for how enterprises operate.

We’ve seen this pattern before:

• Datadog turned fragmented telemetry into a unified source of observability

• ServiceNow started with IT tickets and became the backbone of enterprise ops

• PagerDuty began with alerting and expanded into incident response automation

We believe there’s a similar opportunity here: Start with support. Expand into the operational core. Own the live product knowledge layer.

If You’re Building This, Let’s Talk

If you’re working on code-aware agents, production-grounded AI infrastructure, or anything that helps customer-facing teams operate with real technical context, I want to hear from you. Reach me at priyanka@work-bench.com 📧

Priyanka 🌊

…

I’m a Principal at Work-Bench, a Seed stage enterprise-focused VC fund based in New York City. Our sweet spot for investment at Seed correlates with building out a startup’s early go-to-market motions. In the cloud-native infrastructure and developer tool ecosystem, we’ve invested in companies like Cockroach Labs, Run.house, Prequel.dev, Autokitteh and others.