# Webinar: Inside the Agentic Enterprise > Most enterprises get almost nothing back from AI. MIT says 95% see no real ROI. Models are commoditised now, the advantage is what you build around them. **Published:** 14 July 2026 **Last Updated:** 14 July 2026 **Source:** [https://www.elsewhen.com/blog/agentic-enterprise/](https://www.elsewhen.com/blog/agentic-enterprise/) --- [BUTTON: Download the Slides](https://6082761.fs1.hubspotusercontent-na1.net/hubfs/6082761/2026/Inside%20the%20agentic%20enterprise%20webinar%202026.pdf)## Inside the Agentic Enterprise: Why Most AI ROI Stalls and What Works Most enterprises are spending heavily on AI and seeing very little back. MIT puts it bluntly: 95% of organisations report no meaningful ROI from their AI investments. The National Bureau of Economic Research found that 80% detect no real impact on either employment or productivity. Even adoption inside the tools people already pay for is thin. Only 3.3% of Microsoft 365 users pay for Copilot, according to Windows Central. In this webinar, Elsewhen Co-Founder and Chief Product and Strategy Officer Leon Gauhman set out why this is happening, and what the firms getting real value are doing differently. ## The model is no longer the hard part Start with the thing everyone fixates on: the model. Frontier LLMs from OpenAI, Anthropic, Google, Meta and the Chinese labs now cluster within a few points of each other on aggregate benchmarks. Sam Altman talks about intelligence becoming a utility you buy on a meter. The price of that utility is collapsing fast, with the cost per token for a given level of capability falling between 9x and 900x a year, according to Epoch AI. So the model is a commodity. Everyone has the same magic box. The interesting question is no longer which model you pick. It is what you build around it. That reframes the cost problem too. The early instinct, pouring tokens over the org and hoping for value, turned out to be a bad idea. One firm reportedly burned through $500 million in a month after failing to set usage limits. But runaway spend is a scalability problem, not a model problem. It comes down to context quality, model routing, harness design and measurement. Get those right and the economics work. ## SaaS was built for a world where software was scarce That world is gone. Since late last year, coding agents crossed a threshold. Stripe now merges over 1,300 pull requests a week with no human-written code. Uber runs 1,800 AI code changes a week through its internal agent. Spotify's best developers have not written code by hand since December. When software can be generated on demand, the logic of renting pre-built, one-size-fits-all tools breaks down. Markets noticed: Reuters reported roughly $1 trillion wiped from software market caps as investors repriced the sector. The shift is away from rented tools and towards generated capability, with enterprises writing and owning much more software, not less. Gartner sizes overall IT spend at $5.6 trillion and enterprise software at $1.2 trillion. The operating model itself is becoming code, something you can version, test, roll back and own. Kirkland & Ellis, the world's highest-grossing law firm, is spending $500 million building its own AI rather than renting someone else's. ## The glue is the work Here is the core argument. A company is a collection of workflows, and a workflow is just an algorithm: a series of steps to reach a goal. Today the glue that connects those steps is people, re-keying data between an ERP, a CRM, spreadsheets and email, holding the real context in PDFs and in their heads. The agentic enterprise replaces that glue. The point is not to slap a chat box on every screen. Microsoft quietly shelved its plan to push Copilot across Windows for exactly this reason. The work is connecting models, knowledge, workflows and people into something that does a real job: connected to your systems, grounded in your data, following your process, checking its own output. In the enterprise, your context is the moat. ## Where this goes None of this points to fewer people. Jevons paradox is already visible in software engineering, where cheaper code has expanded demand for engineers, not shrunk it. The human role shifts left, from executing work to deciding what to build, inventing new products and verifying the output of agents. And we are early: most of the world has barely touched this technology. For Elsewhen, the practical answer to the glue is the AI Productivity Platform: five accelerators, Business Context, Headless Agents, Orchestrator, Skills Factory and Generative UI, built on your own stack so you own the IP. It is not a SaaS product. It is a set of components, distilled from patterns we see repeatedly in real work, that our AI Squads use to take a use case from prototype to validated value in four weeks. Intelligence is cheap now. What you build around it is the whole game. See what we can build in four weeks. [Get in touch.](https://www.elsewhen.com/contact)