- HEADLINE + INTRO
OpenAI’s Enterprise Reality Check: Why GenAI Still Hasn’t Rewired Business
When OpenAI’s COO says AI hasn’t really penetrated enterprise business processes yet, he’s puncturing a lot of conference-stage bravado. It also quietly resets expectations: the generative AI boom may be minting revenue, but it hasn’t yet transformed how most companies actually work. This gap between hype and real operational change is where the next wave of winners – and a fair number of losers – will emerge. In this piece, we’ll look at what Brad Lightcap’s comments really signal, how OpenAI’s new Frontier platform fits in, and what all this means for incumbents, consultants, and especially for enterprises in Europe and beyond.
- THE NEWS IN BRIEF
According to TechCrunch, OpenAI COO Brad Lightcap used the India AI summit in New Delhi to deliver a candid assessment: despite powerful models like ChatGPT, AI has not yet deeply embedded itself into core enterprise processes.
The remark comes shortly after OpenAI launched OpenAI Frontier, a platform for enterprises to build and manage AI agents. Lightcap described Frontier as an experiment in bringing AI into the “messy and complex” parts of organisations, and said its success will be measured by business outcomes rather than classic “seat license” metrics. Pricing has not yet been disclosed.
TechCrunch reports that OpenAI ended 2025 with more than $20 billion in annualised revenue, citing CFO Sarah Friar, and that Lightcap says the company is still struggling with excess demand. OpenAI has also partnered with major consultancies such as BCG, McKinsey, Accenture and Capgemini to accelerate enterprise adoption, and announced new offices and enterprise deals in India, where it claims over 100 million weekly ChatGPT users.
- WHY THIS MATTERS
Lightcap’s comment cuts against the usual Silicon Valley script. Instead of declaring victory, OpenAI is admitting that most enterprises are still in the experimentation phase. That honesty is strategically useful: it reframes the race from “who has the biggest model” to “who can actually change workflows at scale.”
The winners in this phase are not just model providers. Systems integrators and consultancies – the very firms OpenAI just partnered with – are positioned to capture a large slice of the value. Turning a chatbot into a procurement assistant or an AI-powered claims process isn’t a prompt-engineering problem; it’s an organisational design and integration problem. That’s McKinsey and Accenture territory.
Traditional SaaS isn’t “dead,” despite the memes that every business app will be replaced by agents. If anything, SaaS platforms with deep domain models and customer data – think CRM, ERP, HR, ticketing – now sit on extremely valuable real estate. They control the workflows that AI needs to inhabit. OpenAI’s Frontier is, in effect, a bid to become a new orchestration layer across those systems before Salesforce, SAP, ServiceNow or Microsoft Copilot lock that position in.
Lightcap’s focus on measuring outcomes rather than seats is also telling. It hints at a pricing and product strategy closer to “we get paid when your process improves” than “we sell you another dashboard.” If OpenAI can make that credible, it threatens both low-value AI-washing tools and parts of the mid-tier SaaS ecosystem that have thrived on shelfware.
- THE BIGGER PICTURE
This announcement sits at the intersection of several converging trends.
First, the industry is pivoting from chatbots to agents and workflows. The initial wave of generative AI was largely about natural language interfaces. The next wave is about systems that can take actions across multiple tools – filing tickets, updating CRM records, generating and executing code, and coordinating with humans. OpenAI Frontier, Anthropic’s new enterprise plugins, and similar offerings all reflect this same shift.
Second, we’ve seen this movie before with robotic process automation (RPA). A few years ago, vendors like UiPath and Blue Prism promised “digital workers” that would automate back-office tasks. Adoption did grow, but it hit hard limits: brittle scripts, complex exception handling, and governance issues. Agentic AI risks repeating those mistakes unless it learns from RPA’s playbook: strong monitoring, process discovery, and realistic expectations about what can be automated.
Third, the competitive field is much wider than OpenAI plus a few US rivals. Enterprises are increasingly evaluating open-source and self-hosted models for workloads where data sensitivity, latency, or cost matter more than raw capability. In finance, healthcare, and the public sector, this is already standard. If Frontier is too closed or too tightly coupled to OpenAI’s own models, it could push serious buyers toward more open stacks.
Zooming out, Lightcap’s remarks confirm that we’re still early in the S-curve of enterprise adoption. The consumer side of generative AI leapt ahead because the friction was low: open a browser, type a question. In enterprises, the friction is organisational: compliance, integration, unions, legacy systems, and very real concerns about jobs. The industry narrative is racing ahead of that reality.
- THE EUROPEAN / REGIONAL ANGLE
For European enterprises, Lightcap’s admission is almost reassuring: if you feel behind, most of the world is too. The constraint is less about vision and more about risk and regulation.
The EU’s regulatory stack – GDPR, the Digital Services Act, the upcoming AI Act, and data transfer rules shaped by Schrems II – means European CIOs have to think harder about where data flows, how models are trained, and how decisions are explained. That slows experimentation, but it also forces better governance from day one.
This creates an opening for European players. Cloud providers like OVHcloud, Deutsche Telekom, or Scaleway, and model companies such as Aleph Alpha or Mistral AI, can position themselves as “AI that plays nicely with EU rules.” If OpenAI wants Frontier to become the process layer in Europe, it will need credible answers on data residency, model customisation, and auditability – and probably local partnerships beyond the global consulting giants.
European enterprises also skew heavily towards sectors like manufacturing, automotive, healthcare, and public services, where process complexity and safety requirements are high. Agentic AI in a factory or a hospital is a much harder sell than in a marketing department. Expect early, deep deployments in lower-risk domains: multilingual customer support, document-heavy compliance tasks, internal knowledge search, and software development.
Finally, while Lightcap highlighted voice as a growth vector in India, the multilingual nature of Europe makes robust speech and translation capabilities equally strategic here – particularly for cross-border organisations operating in a dozen languages.
- LOOKING AHEAD
Over the next 12–24 months, the story of enterprise AI will be less about new model launches and more about boring, specific questions: Which processes moved the KPI needle? Which deployments survived contact with legal, HR, and works councils? Where did automation quietly fail?
Expect a wave of narrow but deep use cases to harden first: contact centres, IT support, sales enablement, software engineering, and internal research. In each, agents will be judged on tangible metrics – resolution time, error rates, revenue per rep, code throughput – not on how human they sound in a demo.
For OpenAI, the fate of Frontier will hinge on three things. First, how well it plugs into the existing enterprise stack: connectors, identity, permissions, and logging are not glamorous but are non-negotiable. Second, whether the consulting partnerships translate into repeatable “playbooks” that mid-market firms can afford, not just bespoke mega-projects for Fortune 500s. Third, how it navigates the growing tension between closed, vertically integrated platforms and enterprises’ desire to avoid lock-in.
Unanswered questions remain: Will regulators treat powerful agents differently from today’s chat interfaces? How will labour markets respond as some tasks are clearly automated away? And perhaps most crucially, will enterprises have the courage to redesign processes around AI, rather than just sprinkling a chatbot on top of the old way of working?
- THE BOTTOM LINE
OpenAI’s candid admission that AI hasn’t yet penetrated enterprise processes is less a confession of weakness than a sign that the real contest is only starting. The next phase will be won not by the flashiest demos, but by those who can turn models into measurable business outcomes under real-world constraints. For technology leaders, the question is no longer whether to “try” generative AI, but which specific workflows you are prepared to rebuild around it – and how quickly you can move before that opportunity passes to your competitors.
