# From tools to industries Author: Juan Segundo Hevia Original publication: 2026-05-27 Spanish original: https://estrategiabyaleph.substack.com/p/estrategia-139-la-nueva-fase-de-la English URL: https://elcontemplador.github.io/estrategia-english/essays/139/ Status: Published translation This is a translation of the original Spanish essay published on 27 May 2026. Its claims, examples and forecasts retain that historical context. English publication: 2026-09-29 *Editorial introduction:* This week we have the privilege of publishing an article by [Juan Segundo Hevia](https://www.linkedin.com/in/juanse-hevia/), Executive Secretary and Research Director of the Observatory of AI, Innovation, Society and Government at Universidad Austral in Argentina. His work lies precisely at one of the intersections that interests this newsletter most: how artificial intelligence is transforming the economy, institutions, innovation and governments’ ability to understand — and anticipate — technological change. In his article, [Juanse](http://juansehevia.com) analyses a decisive shift: the major AI companies are no longer competing only to offer better tools, but to enter the workflows of entire industries. Particularly relevant reading for governments, regulators and organisations that need to understand AI not only as a technology, but as a new economic and institutional infrastructure. *Archive note: launches, company strategies and interpretations retain the article’s publication context of 27 May 2026.* The week of 19 May concentrated more AI announcements than any other week so far this year. At [Google I/O 2026](https://blog.google/innovation-and-ai/sundar-pichai-io-2026/), Google unveiled [Gemini 3.5 Flash](https://developers.googleblog.com/all-the-news-from-the-google-io-2026-developer-keynote/), its new agentic platform [Antigravity 2.0](https://blog.google/innovation-and-ai/technology/developers-tools/google-io-2026-developer-highlights/) and [Gemini Spark](https://blog.google/innovation-and-ai/technology/developers-tools/google-io-2026-developer-highlights/), a personal agent that operates continuously across Workspace. Days earlier, OpenAI had launched its [Deployment Company](https://openai.com/index/openai-launches-the-deployment-company/) with more than $4 billion in backing, while Anthropic had launched [financial agents](https://www.anthropic.com/news/finance-agents) and an [expanded partnership with PwC](https://www.anthropic.com/news/pwc-expanded-partnership) to certify thirty thousand professionals in the use of Claude. To understand these moves as a system, rather than a succession of news stories, we need a map. ## The paradox Two years ago, [Anthropic](https://www.anthropic.com/) and [OpenAI](https://openai.com/) sold models. Today they sell products for vertical markets: design, code, operational productivity, and they are already beginning to enter entire industries such as law, healthcare, education and creative production. What these companies accumulated was not simply computing power, but something scarcer and harder to replicate: **the traces left by millions of users explicitly stating what they want to achieve**. For anyone designing public policy or regulating strategic sectors, the question is this: how do you move from a general-purpose model to the systematic colonisation of entire verticals? ## Growth and the contrast with Google The use of mass-market products such as ChatGPT creates a feedback cycle: model use generates data that improve the model and its derivative products, and those products in turn attract more users. So far, nothing distinguishes this from the classic cycles of the previous era: Google’s search, [Amazon’s](https://www.amazon.com/) e-commerce or [Meta’s](https://about.meta.com/) feed likes. The difference, however, is qualitative. The data Anthropic and OpenAI capture are much more valuable than mere clicks, impressions or browsing patterns. Minute by minute, they collect **intentional conversations**: moments when users explicitly state what they want to achieve, how they iterate, what frustrates them and where they give up. Every marketing department’s dream. That intention, aggregated at scale, is probably the most valuable asset in today’s digital world, because it becomes an instruction manual for capturing unmet demand. Google has its own feedback process too, and I/O 2026 made it visible: [Stitch](https://stitch.withgoogle.com/) for design, now with [Gemini 3 and real-time design mode](https://blog.google/innovation-and-ai/models-and-research/google-labs/stitch-updates/); [Flow](https://labs.google/flow/about) for generative images and video, the successor to [Whisk](https://workspaceupdates.googleblog.com/2026/03/whisk-is-moving-to-flow-on-april-30-2026.html), discontinued on 30 April 2026; [Gemini](https://gemini.google.com/) integrated throughout Google [Workspace](https://workspace.google.com/), now with [Gemini Spark](https://blog.google/innovation-and-ai/technology/developers-tools/google-io-2026-developer-highlights/), an agent operating continuously across email, documents and spreadsheets. Add to these its flagship generative products: [NotebookLM](https://notebooklm.google.com/), a foothold in knowledge management, and the new version 2.0 of its agent-based software development environment, [Antigravity](https://blog.google/innovation-and-ai/technology/developers-tools/google-io-2026-developer-highlights/). Google is playing a different game from its competitors, protecting the ecosystem of applications it already owns and monetises. It defends itself by implementing AI to expand the products and services for which it already charges. Anthropic and OpenAI have no suite to defend. They therefore take an **offensive** stance, to the point that OpenAI created DeployCo, [a separate company for implementing AI in businesses](https://openai.com/index/openai-launches-the-deployment-company/), with more than $4 billion in backing and partners including TPG, Goldman Sachs, Bain and McKinsey. Anthropic, meanwhile, formed its own [AI services company with Blackstone, Hellman & Friedman and Goldman Sachs](https://www.anthropic.com/news/enterprise-ai-services-company), reinforced a few days later by an [expanded partnership with PwC](https://www.anthropic.com/news/pwc-expanded-partnership) to deploy Claude to hundreds of thousands of professionals and certify 30,000 of them in the model’s use. In short, they are free to target any vertical where the data show density, without cannibalising their own businesses. That asymmetry defines each player’s vertical reach. ## The three stages of vertical expansion From within, the cycle operates in three moves: 1. **Capture.** The model observes millions of real conversations through its technical interface, the API, and the direct-use product — [Claude.ai](https://claude.ai/) or [ChatGPT](https://chatgpt.com/). It records not just what people ask, but how they iterate, which solutions they accept and where they give up. This is the most precise X-ray ever produced of the problems people are trying to solve with software. 2. **Pattern.** The laboratory’s teams identify recurring, high-demand tasks — jobs to be done: preparing a legal opinion, designing a presentation, rewriting a fragment of code, planning a lesson or drafting a public-policy memo. Every pattern with sufficient density is a candidate for a vertical. 3. **Vertical.** The laboratory releases a specialised product that internalises the pattern: Claude Skills — [Design](https://www.anthropic.com/news/claude-design-anthropic-labs), [Legal](https://es.qz.com/antropic-claude-herramientas-de-ia-legal-para-firmas-de-abogados-051226), [small businesses](https://www.anthropic.com/news/claude-for-small-business) and others — and personalised [Codex](https://openai.com/index/work-with-codex-from-anywhere/) agents. Even connectors to third-party software through the standard these laboratories themselves created: [MCP](https://modelcontextprotocol.io/), the Model Context Protocol, an open standard that lets AI models connect in an orderly way to external applications. The vertical product generates even more structured data than the general model, feeding the next iteration of the cycle. Product development accelerates and the move into a vertical — design, law or consultancy — gains strength. ## The two waves of colonisation This engine has not unfolded uniformly: it advances in waves, each with a different logic. **The first wave involved adjacent productive workflows.** Design — [Claude Design](https://www.anthropic.com/news/claude-design-anthropic-labs), Stitch and native image functions in ChatGPT; code — [Codex](https://openai.com/index/codex-for-almost-everything/), [Claude Code](https://www.anthropic.com/claude-code), OpenAI’s and Anthropic’s integrated environments; knowledge management — NotebookLM, corporate Custom GPTs and [ChatGPT connected to bank accounts for personal financial planning](https://openai.com/index/personal-finance-chatgpt/). In this wave, AI replaces **tools**. [Figma](https://www.figma.com/) loses ground to native generators, extensions for development environments such as [Visual Studio Code](https://code.visualstudio.com/) compete with integrated assistants, and wikis such as [Stack Overflow](https://stackoverflow.com/) or [Wikipedia](https://www.wikipedia.org/) give way to conversational interfaces that answer questions on any subject. Friction is low because users are already digital and the value chain is short. **The second wave, now under way, involves entire industries.** [Law](https://es.qz.com/antropic-claude-herramientas-de-ia-legal-para-firmas-de-abogados-051226), with vertical assistants trained on case-law and contract corpora; [finance](https://www.anthropic.com/news/finance-agents), with agents for banking, fund management, know-your-customer compliance and accounting close; health, with diagnostic support, clinical management and medical documentation; education, with adaptive tutoring and assessment; and the [creative industries](https://www.anthropic.com/news/claude-for-creative-work), with native connectors to [Adobe](https://www.adobe.com/), [Blender](https://www.blender.org/), [Autodesk](https://www.autodesk.com/) and [Splice](https://splice.com/). Here, it is no longer a tool being replaced: the model enters a sector’s actual workflow, with its practices, regulations and chain of participants. Friction is high, but when capture occurs it is structural. This process may have the most revolutionary implications for how millions of people work day to day. AI ceases to be a copilot generating information for us and becomes an entity to which we can delegate our daily tasks. Tasks it progressively performs faster and better than we do, at a fraction of the cost, because the usage data feeding back into it relate directly to the industry in which it is being used. And this second wave is one Google is unlikely to be able to pursue with the same freedom. Embedding Gemini in Blender or in a hospital management system means reorganising industries in which Google has no foothold; it lacks any guarantee that this expansion will feed back into the business it already monetises. **I/O 2026 confirmed this by omission:** none of the major bets Google showed that week — Antigravity, Spark, Gemini 3.5 Flash, smart glasses and Android XR — targeted a particular regulated industry. All reinforce the horizontal perimeter the company already dominates, or seek to open a new one from scratch, consumer hardware, but none enters a vertical through its front door. **Anthropic and OpenAI face no such dilemma: their only loyalty is to the usage pattern.** ## Implications for politics and government **What is at stake is no longer the software market, but regulated industries with their own regulatory frameworks:** professional bodies, sectoral ministries, education authorities and health agencies. AI does not arrive as a new tool; it arrives as infrastructure, reorganising the sector’s value chain and redistributing power among existing actors. This unsettles sectoral regulators accustomed to regulating finished products rather than horizontal platforms becoming vertical providers. For governments with ambitions for technological sovereignty, the consequence is uncomfortable: training a national model is no longer enough on its own. **The critical asset is the trace of sectoral usage: who observes how judges, doctors, teachers or engineers work within a jurisdiction.** Without vertical expansion of your own, you lose even if you have a model. The window, as these pages often remind us, is narrow. The first wave replaced tools. The second is replacing industries. Anyone regulating the tool rather than the infrastructure will arrive late to a conversation already taking place in another language. [Juan Segundo Hevia](https://www.linkedin.com/in/juanse-hevia/) *Executive Secretary and Research Director of the Observatory of AI, Innovation, Society and Government at Universidad Austral* [juansehevia.com](http://juansehevia.com) [![estrategIA infographic explaining vertical expansion from tools into industries, credited to Juan Segundo Hevia.](https://elcontemplador.github.io/estrategia-english/assets/images/e6dd3150-c9e6-4faa-ad64-f9d7cdf0a333_864x1821.png)](https://substackcdn.com/image/fetch/$s_!sE_D!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe6dd3150-c9e6-4faa-ad64-f9d7cdf0a333_864x1821.png) *English transcription of the accompanying infographic, branded estrategIA by ALEPH: “From tools to industries — The second AI wave no longer replaces software: it reorganises entire sectors”.* “The critical asset is no longer just the model: it is the trace of sectoral usage.” — estrategIA. **Central idea:** OpenAI and Anthropic move from selling models to capturing verticals: design, code, law, finance, health, education and creativity. Google defends its horizontal ecosystem. **The engine of vertical expansion:** 1. Capture — conversations, iterations, frustrations and abandonment. 2. Pattern — recurring jobs to be done with high demand. 3. Vertical — specialised products that generate more structured data. | Two strategies | Details | | --- | --- | | Google | Protects Workspace, Search and Android; integrates Gemini into its own products; plays horizontally. | | OpenAI + Anthropic | No suite to defend; DeployCo, Claude, PwC and partnerships; an offensive game targeting verticals. | **The two waves:** wave one, workflows — design, code and knowledge; AI replaces tools. Wave two, industries — law, finance, health, education and creativity; AI enters the sector’s actual workflow. **Implications for governments:** 1. Regulate infrastructures, not only tools. 2. Build their own traces of sectoral usage. 3. Prepare ministries and regulators for vertical platforms. Closing text: “The first wave replaced tools. The second is replacing industries.” Credit: Juan Segundo Hevia · Observatory of AI, Innovation, Society and Government · Universidad Austral.