TL;DR
- Conduct raised $60M in June 2026 on a single premise: AI agents can only act on systems they actually understand.
- Decades of custom enterprise configurations have made most business software opaque — even to the engineers who run it.
- Companies like Fraport, DHL, and Daimler Truck are already seeing 30%+ acceleration in transformation projects once a legibility layer is in place.
- For enterprises evaluating AI, the real question isn't "which tool?" — it's "are our systems even ready for an agent to read?"
The Uncomfortable Truth Behind Every AI Pilot
There's a scene playing out in enterprise IT departments across the world right now. A leadership team signs off on an AI agent initiative. The vendor demos look spectacular. The board is excited. Then the implementation team opens the hood of the actual software environment — and discovers something that resembles less a finely tuned engine and more a decade's worth of duct tape, custom middleware, and undocumented integrations held together by institutional memory and optimism.
This is the wall AI agents keep hitting. And it's not a problem of model intelligence. It's a problem of software legibility.
What $60 Million Is Actually Buying
In June 2026, Conduct closed a $60M Series A co-led by Index Ventures and ICONIQ, with SAP joining as a strategic investor. The bet is straightforward, if underappreciated: before any AI agent can act on an enterprise system, it first has to understand that system.
Conduct's CEO put it plainly — "An agent can only act on a system it understands." Seven words that quietly invalidate a significant portion of enterprise AI roadmaps currently in flight.
What makes this funding round notable isn't just the size. It's the strategic validation. SAP, a company that has spent decades embedded in the messiest layers of enterprise software, is backing the idea that legibility is infrastructure. That's not a casual endorsement.
Companies like Fraport (the operator of Frankfurt Airport), DHL, and Daimler Truck are already working within this framework — and reportedly seeing 30%+ acceleration in transformation timelines once a legibility layer is established. These aren't startups running greenfield experiments. These are large, complex organizations with exactly the kind of sprawling, historically-accrued software environments that make AI agents stumble.
Why Enterprise Software Is So Hard to Read
To understand why legibility matters, it helps to appreciate just how opaque enterprise software landscapes tend to become over time.
Most large organizations don't run clean, well-documented systems. They run living archaeological sites — layers of ERP customizations from 2009, integrations built by contractors who left in 2015, business logic encoded in places no one thinks to look, and critical workflows that exist only because someone named Gerhard figured it out and never wrote it down.
This isn't negligence. It's the natural entropy of any complex system that has had to adapt to real business conditions over time. But it creates a fundamental problem for agentic AI: an agent that can't interpret what a system does — and why — can't reliably act on it without causing unexpected downstream chaos.
As one observer noted on X:
"This perfectly captures the reality of AI-powered development. Building the app is only the beginning—security, governance, access control, and deployment are where enterprise readiness is truly defined. Great breakdown of the gap between a demo and a production-grade application. 🚀"
— @VictoriaBlddd
That gap between demo and production is exactly where legibility lives. A well-behaved agent in a controlled demo environment and a well-behaved agent navigating a real enterprise stack are, in many cases, very different propositions.
The Agentic Shift Is Real — But It Has Prerequisites
The broader industry conversation has moved decisively. We are, by most credible accounts, past the era of AI as a chatbot novelty and into the era of governed autonomy — agents that plan, reason, and execute across enterprise systems without constant human hand-holding.
"The Agentic Shift: Why 2026 is the Era of Governed Autonomy. We have officially moved past the 'chatbot era.' June 2026 marks a definitive shift toward Agentic AI — autonomous systems that reason, plan, and execute tasks across enterprise stacks without constant human intervention."
— @PawanConnect
This shift is real. But governed autonomy requires something to govern against — a clear, interpretable map of what the systems contain, how they connect, and what the business logic actually demands. Without that foundation, "agentic AI at scale" is less a capability and more a liability.
This is also showing up in adjacent spaces. Platforms like Poolside — built by veterans from GitHub and DeepMind — are developing enterprise-grade agentic systems specifically designed for complex, long-horizon software engineering tasks. The community reaction to this kind of infrastructure-first AI has been telling:
"Finally, an AI coding platform that enterprises can actually run inside their own security boundary with real governance."
The emphasis on governance, auditability, and security boundaries isn't incidental. It reflects the same underlying insight driving Conduct's funding: the precondition for capable AI action is a legible, trustworthy environment.
Even in less traditional contexts, the instinct toward structured AI accountability is gaining momentum. A project submission from a recent hackathon illustrated the point:
"RecallOps — A human-owned recall command system that uses AI to aggregate data and compute evidence, keeping legal decisions accountable and audit trails immutable."
— @Ggudman1
The pattern repeats: AI doing more, but only within frameworks where humans can audit, override, and understand what's happening.
What This Means for Swiss Enterprises
For Swiss organizations currently evaluating AI implementations — and there are many — the Conduct story reframes the evaluation criteria in an important way.
The question most procurement teams are asking is: "Which AI tool should we adopt?"
The question they should be asking first is: "Can an AI agent actually read our current software landscape?"
This isn't an argument against moving forward with AI. It's an argument for sequencing correctly. Procuring a sophisticated AI agent and pointing it at an undocumented, heavily customized ERP environment is roughly equivalent to hiring a brilliant new analyst and handing them a filing cabinet with no labels, no index, and three drawers that are mysteriously locked.
The analyst isn't the problem. The filing cabinet is.
This is where specialized IT architecture partners earn their value in 2026. Firms positioned at the intersection of IT architecture and AI deployment — capable of auditing a software landscape for AI-readiness before agents are deployed — are offering something more durable than implementation support. They're offering the prerequisite that makes implementation actually work.
The Real IT Revolution of 2026
The headline narrative around AI and enterprise IT has often centered on developer displacement — the idea that AI will replace engineering teams. The more honest story emerging in 2026 is subtler and, frankly, more interesting.
It's not about replacing developers. It's about making systems legible enough that AI can work alongside them.
That requires real expertise: understanding enterprise software architecture, mapping integrations, surfacing undocumented business logic, and building the kind of interpretable layer that agentic AI needs to operate safely and effectively.
The companies getting ahead of this — like Fraport, DHL, and Daimler Truck — aren't just buying AI tools. They're investing in the scaffolding that makes those tools meaningful. And the $60M backing Conduct suggests that investors are now pricing this scaffolding as core infrastructure, not nice-to-have consulting.
For any enterprise approaching an AI agent initiative in the second half of 2026, the smartest first move might not be a vendor selection. It might be an honest audit of whether your systems are ready to be understood.
Because an agent can only act on a system it understands. And understanding has to be built before anything else can happen.
Interested in assessing your organization's AI-readiness before the agents arrive? Reach out to explore what a software legibility audit looks like in practice.
Published in Stream · Dispatch #434 · June 20, 2026 · 7 min read.
Reply to paolo@mont3.ch - every email gets a human answer within 24h.