The most interesting work in artificial intelligence right now isn’t happening in the model itself. It’s happening in the unglamorous layer underneath: the systems that grant or deny access, the checks that decide whether an output can be trusted, the infrastructure that lets a voice agent understand a customer in real time. The San Francisco Tribune put together a list of companies working on exactly that problem set ahead of HumanX Amsterdam, part of the wider HumanX Europe series, running September 22 through 24 at RAI Amsterdam.
The Plumbing Problem
Every enterprise AI rollout eventually runs into the same wall: the systems that hold the real data and the real business logic were never designed to give either one up easily. Unfold is building for exactly that moment. Still in stealth, the company says it can make closed, undocumented, or vendor-locked systems readable without an API, without documentation, and without needing the vendor in the loop, all while operating inside the customer’s own environment with read-only access and full auditing by default. Its underlying argument is hard to dispute: business logic that stays inaccessible can’t be modernized, and it certainly can’t be reasoned over by AI.
Kensho Technologies solves a related but different version of the same problem. As S&P Global’s innovation engine, it connects large language models and AI agents to trusted financial and business data while structuring that proprietary data for machine learning and generative AI workflows. In finance, where data quality is everything, that pairing of domain expertise with AI infrastructure has become a real part of the enterprise stack.
Nine Companies, One Direction
Reliability is its own category of problem, and causaLens has built its entire platform around closing the gap between an impressive agentic demo and something that survives contact with production. Its Digital Knowledge Workers automate repetitive knowledge tasks using multi-agent architectures, supported by pre-built Blueprints, a Factory for customizing new workers, and a governance layer called the System of Work. Causal reasoning and human-in-the-loop controls sit at the center of the approach, with the goal being measurable outcomes instead of a good-looking pilot.
Trendium takes on the governance side directly. Its ContextGuard platform gives enterprises visibility and control over their AI infrastructure, and its AI Enablement Program helps engineering teams adopt AI coding agents responsibly, all organized around four areas: visibility, control, enablement, and compliance.
Voice AI shows up three times on this list, and each company is solving a slightly different piece of it. AssemblyAI supplies the developer-facing infrastructure, with APIs for transcription, contextual understanding, and real-time agentic workflows. Speechmatics goes after low-latency recognition across more than 55 languages and multiple speakers, deployable in the cloud, on-premises, or on a device, serving everything from legal transcription to live captioning. Otter AI is stretching furthest beyond the transcription category itself, positioning its platform as a conversational knowledge engine that turns meetings into searchable, queryable organizational memory.
The remaining two companies are applying AI to problems outside the usual enterprise-software conversation entirely. Nuritas uses its proprietary Nuritas Magnifier platform to discover and validate peptides from nature, having already identified more than 8 million of them and built a library of known functionalities, moving discoveries from computational prediction to clinical validation in a fraction of the time that process usually takes. PhotoRoom is building visual production infrastructure for e-commerce, offering batch editing, automated quality assurance, and brand controls to sellers of every size, with a strong focus on keeping AI-generated product images faithful to what’s actually being sold.
Where This Leaves Enterprise Buyers
None of these nine companies are chasing headlines with a bigger, flashier model. They’re solving the problems that determine whether an AI system that works in a demo will still work six months into production, with real data, real users, and real consequences. HumanX‘s own event structure, including its VentureConnect and SolutionBridge programs, is built around exactly that kind of conversation between startups and the enterprise buyers who need these problems solved.
