The Missing Half of Design

Across 116 AI companies in our dataset there are 9,281 active job postings. 2,530 of those — 27.3% of all hiring — sit in engineering. 195 sit in the design function. That's 13 engineers being hired for every designer, across an industry building the products that are supposed to reshape how software works.

The second reading of the same data complicates the first. In a separate function, 1,303 active postings — more than six times the entire design headcount — are for Forward Deployed Engineers, Solutions Architects, Customer Success Managers, and Engagement Managers. These job descriptions use the vocabulary that used to belong to design. 49.8% of them mention adoption. 26.5% mention user or customer journeys. 23.5% mention discovery. 10.4% mention stakeholders. On every one of those axes, deployment roles track product designers and product managers more closely than they track engineers, who almost never speak in that vocabulary at all.

The 13:1 ratio is a real finding. It is also not the whole finding. The design function at AI companies has not disappeared. It has bifurcated. One half — product and interface design — is structurally understaffed relative to engineering, and it is understaffed even at companies that demonstrably have software products to design. The other half — service design, the mapping of workflows and stakeholders and adoption paths — has moved out of the design function entirely and into an adjacent, much larger function that AI companies have built from scratch.

Here is what the missing half of design looks like.

The aggregate, and what disaggregation does to it#

Engineering is 2,530 of 9,281 active postings. The design function — product designers, brand designers, motion designers, UX researchers, design systems designers, and the long tail of creative and visual roles — is 195. At the aggregate level this is 13.0 engineers per designer across AI companies.

The aggregate is an average of very different populations. Once companies are segmented by what they build, the ratio fans out across an order of magnitude.

Companies building only applications — the 42 companies in our dataset closest to end users — hire 7.5 engineers per designer. That is tight for AI but still loose by the benchmarks of consumer software a decade ago. Alex Schleifer, former chief design officer at Airbnb, grew design headcount from roughly 35 to over 600 during his tenure, tracking engineering's growth. IBM under Phil Gilbert publicly named a 1:8 designer-to-developer target in 2012 after concluding that its existing 1:72 ratio was unworkable. In 2026, the segment of the AI industry closest to consumers is operating roughly halfway between IBM's starting point and IBM's target.

The ratio widens from there. Pure Models companies — the nine early-stage labs that only train models and do not yet ship infrastructure or applications — run 7.7:1. The combined Models+Apps segment (24 companies, including Anthropic's consumer work, the frontier video and image model companies, and applied-science firms like Recursion and Isomorphic Labs) runs 9.1:1. Pure Infrastructure runs 31.8:1. Companies building across all three layers of the stack run 41.5:1.

The counter-case is real but narrow. Linear hires 1.7 engineers per designer. Lovable, 2.3. Runway, 2.7. Figma, 3.7. A small cluster of application-layer AI companies — mostly consumer-facing, mostly craft-forward — runs ratios that look like Airbnb or Stripe at similar stages. But they are the exception. The modal AI company in the dataset is closer to Databricks (39.2:1) or Waymo (53.0:1) or CoreWeave (81.0:1) than to any of them.

It is not a UI-surface problem#

The first objection to the 13:1 number is that infrastructure and model companies do not need designers because they do not have user interfaces. The objection is partially true and mostly false.

Product management hiring is a workable proxy for "this company has a software product." PMs do not get hired at companies that do not have surfaces to manage. If a company is hiring product managers but not designers, the UI exists and the design staffing does not.

Across the 22 Pure Infrastructure companies in our dataset, 8 are actively hiring product managers. Of those 8, 5 are hiring zero designers. At Models+Apps, 3 of 3 PM-hiring companies have no active design hiring. At Models+Infra, 2 of 7. Across the full dataset, fourteen companies are hiring PMs and zero designers right now.

The list is specific: MongoDB (418 active postings, 8 PMs, 123 engineers, 0 designers) — Atlas has one of the most heavily-used consoles in enterprise infrastructure. Mistral AI (149 postings, 6 PMs, 35 engineers, 0 designers) — ships Le Chat to consumers and La Plateforme to developers. Cohere (115 postings, 3 PMs, 45 engineers, 0 designers) — runs a developer dashboard and a consumer chat product. Lambda (30 postings, 1 PM, 10 engineers, 0 designers) — runs a GPU cloud console used by thousands of teams. Snorkel AI (45 postings, 2 PMs, 14 engineers, 0 designers) — ships a data-labeling UI that is the core of its product. RunPod (19 postings, 1 PM, 6 engineers, 0 designers) — pod management and templates. Cerebras Systems (96 postings, 2 PMs, 49 engineers, 0 designers) — training platform UI.

The "no UI to design" explanation holds for a narrow cohort: Cerebras and Graphcore (primarily silicon), Figure AI (robotics hardware, with industrial designers filed under a different role), Physical Intelligence, Recursion, and Isomorphic Labs (where the product is a pipeline rather than a UI). Five or six of the sixteen companies in the data with 20+ jobs and zero designers plausibly sit in this category. The remaining ten have products, users, and UIs. They have chosen not to staff design against them.

When infrastructure companies do hire designers, they hire them for the consumer surface#

The second refinement: when Pure Infrastructure and Models+Infra companies do hire designers, the titles are specific, and they are not the titles a traditional enterprise software company would hire.

Nebius, the largest example in the dataset with active design hiring in the Pure Infrastructure segment, has four product designer searches: Senior Product Designer, Nebius Console, Senior Product Designer, Token Factory, a general Designer (UX + Visual), and a UX Writer for the Console. These are designers for a developer-facing product.

OpenAI is the cleaner illustration of where the investment is going. Of its eleven active design reqs, every one sits on a consumer or monetization surface — Product Designer, ChatGPT, Product Designer, Codex, Product Designer (Growth), Product Designer (Monetization Platform), Lead Product Designer (Health AI), Product Designer (Business Products). There are no designers being hired for the OpenAI platform console, for the API documentation experience, or for the developer admin flows. DeepMind's hiring is starker: all ten of its active product designer searches are for the Gemini app — Gemini Assistant, GeminiApp iOS, GeminiApp Mobile, GeminiApp Growth and Discovery, GeminiApp Device Experience.

The pattern inverts the traditional enterprise-software design investment. At Salesforce or Adobe or Oracle, a substantial share of the design function historically worked on admin panels, integration tools, and reporting surfaces — the parts of the product that operators interacted with. At AI companies hiring designers, the designers are pointed at the consumer-facing layer almost exclusively. Developer tooling, admin consoles, and the long tail of enterprise UI get built without a dedicated design function.

Two framings of this are possible. The first, common in design-industry commentary, is that AI infrastructure is early enough that the developer-facing surface has not yet attracted design investment. The second is that AI companies have absorbed the assumption that developers do not need designed UIs, because developers can self-serve via APIs and CLI tools. The data cannot distinguish between the two. What it shows is that in April 2026, the enterprise-software design assumption has not made it to the AI side.

Service design, filed under "engagement"#

The design count at 195 is the end of one story. The deployment count at 1,303 is the start of another.

Deployment roles — Forward Deployed Engineers, Solutions Architects, Solutions Engineers, Customer Success Managers, Engagement Managers, Technical Support Engineers, Implementation Specialists, Client Partners, Business Value Consultants — were the subject of an earlier piece in this series, where they were framed as the customer-facing technical bridge that makes AI products work inside customer environments. That framing is correct and incomplete. Read at the language level, the deployment function at AI companies is doing work that in an earlier era of enterprise software would have been split across four different disciplines: service design, business analysis, change management, and implementation consulting.

The evidence is in the job descriptions themselves.

Deployment postings mention discovery in 23.5% of descriptions — more than five times the rate in core engineering. They mention stakeholders in 10.4%, six times the rate in engineering. They mention customer or user journeys in 26.5%, higher than product designers themselves (18.7%). They mention adoption in 49.8%. Change management, operating models, and ways of working appear in 6.8% of deployment postings versus 1.8% in engineering. Product managers, for reference, mention discovery in 27.9% of postings, journeys in 24.3%, and adoption in 47.1% — the deployment vocabulary tracks theirs almost exactly.

The volume is asymmetric in a way that compounds the point. 23.5% of 1,303 is 306 deployment postings that mention discovery work. 26.2% of 107 is 28 product designer postings. There are roughly eleven times as many deployment reqs talking about discovery as there are product designer reqs talking about it.

The individual postings read the same way. Notion's Services Engagement Manager is hired to help customers across onboarding, implementation, workflow transformation, and long-term adoption. The description characterizes the services team as helping organizations redesign how work gets done by embedding Notion as a connective operating layer. The phrase "redesign how work gets done" would not be out of place in a service design consultancy's capabilities deck.

The pattern is not specific to Notion. Hebbia's Forward Deployed Engineer posting describes the role as embedded with strategic customers, building the last mile of the platform for their specific workflows. OpenAI's Forward Deployed Engineer for Life Sciences uses early engagements to define repeatable system patterns, evals, and operating standards for life-sciences environments. Neural Concept's Technical Account Manager runs workflow deep-dives for designers and simulation engineers and leads discovery to craft solution-oriented technical proposals. EliseAI's Engagement Manager works on integrating AI agents deeply into existing workflows in housing and healthcare.

The titles are engineering titles and sales titles. The work, read from the postings, is service design.

Where it gets complicated#

Several things the data does not resolve.

Job descriptions are marketing artifacts. They describe what companies want to attract, not what the resulting employee ends up doing. The vocabulary of discovery and journey-mapping and adoption can appear in deployment postings as enterprise-sales language — signaling seriousness to prospective customers — rather than as actual methodological commitment. A Forward Deployed Engineer at OpenAI who shows up on day one may well spend most of their time writing Python and wiring up evaluations, not facilitating stakeholder workshops. The vocabulary is a starting point for the argument, not proof.

Second, the service-design vocabulary is also generic product-management language. Product managers mention discovery in 27.9% of postings and journeys in 24.3% — numbers nearly identical to deployment roles. It is possible that the deployment function is not absorbing service design work so much as absorbing product management for a particular kind of customer engagement. The two are related but not identical, and the data cannot separate them cleanly.

Third, the design engineer role exists but cannot explain the gap. Fifteen active design engineer postings sit across eleven companies, all of them in the Pure Applications or Infra+Apps segments (Lovable, Cursor, Replit, Ramp, Vanta, Runway, and a handful of others). They do not appear at OpenAI, Anthropic, Databricks, Mistral, Cohere, or any of the Pure Infrastructure companies. Design engineering is a real and growing discipline; it is not what model labs and infrastructure companies are doing instead of design.

Fourth, the design function being compared is partial. The data shows hiring, not headcount. DeepMind's Google-era design org, Databricks' existing design team, and Anthropic's earlier hires are not visible here — only what these companies are adding now. Companies with zero active design reqs may have ten designers already on staff. What the data shows is the direction each company is expanding, and the direction is not toward design.

Fifth, the concentration story matters. Remove the top three hirers — Databricks, OpenAI, and Anthropic — from the engineering count and the aggregate ratio compresses from 13:1 to roughly 11:1. The direction is unchanged; the magnitude softens.

Observations#

The 13:1 aggregate is not driven by UI absence. 62.5% of Pure Infrastructure companies actively hiring product managers are hiring zero designers. They have products. The design staffing is a choice.

When AI infrastructure and model companies do hire designers, they hire them for the consumer surface. The admin panels, developer consoles, and integration tools that traditional enterprise software staffed with design teams are being built without them. The design investment at OpenAI, DeepMind, and Anthropic is almost entirely pointed at ChatGPT, Gemini, and Claude's consumer and business surfaces — not at the platforms underneath them.

The service-design function has been reclassified into deployment. 1,303 active deployment postings — six and a half times the entire design function — use discovery, journey, stakeholder, and adoption language at rates that track product designers and product managers, not engineers. The work that used to be service design, business analysis, change management, and implementation consulting in enterprise software has been consolidated into Forward Deployed Engineers, Engagement Managers, and Solutions Architects.

Design at AI companies has not shrunk. It has bifurcated. Product design is understaffed against engineering by any comparable historical benchmark. Service design has been renamed. The titles are different, the reporting lines are different, the function in the org chart is different. Most of the design work is still happening — much of it is now happening somewhere other than the design function.

The roles and companies in this analysis — the Product Designer at Harvey, the Design Engineer at Lovable, the Services Engagement Manager at Notion, the Forward Deployed Engineer at OpenAI — can be explored at appliedmethods.ai.

All data is from the Applied Methods dataset as of April 19, 2026. Job counts reflect active postings at time of analysis. The segmentation of companies by what they build (models, infrastructure, applications) is based on editorial classification of each company's primary product. Vocabulary analysis of deployment-role descriptions uses case-insensitive regex matches and will overcount postings where a term appears in unrelated context (for example, "discovery call" rather than "discovery research"). The dataset covers 116 AI companies — primarily venture-backed startups and public companies with significant AI operations — and does not cover AI adoption at traditional enterprises. All roles mentioned can be explored at appliedmethods.ai.