Tommy Geoco - The state of the design industry right now

2026-05-19 - 60 min - source - Read full transcript

Rid (host)Tommy Geoco

Key insights

Designers who spend the majority of their workflow vibe coding report higher satisfaction, but adoption skews away from junior and mid-level ICs.

Tommy's state-of-prototyping survey found design engineers, leads, and principals adopt most, followed by non-designer roles like students and researchers, then managers, with general junior/mid-level ICs adopting least - suggesting either a trickle-down in progress or that some teams simply haven't made room for the rest of the team yet.

vibe-coding-adoption

59% of designers have built their own internal tool for their workflow, and dismissing this activity as 'AI slop' misreads what's actually happening.

Most of these tools are narrow, single-purpose fixes - a Vercel brand designer built a tool that auto-applies branding and sizing to marketing blog images so the design team wouldn't have to manually recreate them - and get discarded or shared internally once they've served their purpose.

internal-tool-building

Role labels are compressing toward 'builder' as design, marketing, sales, and brand work all start running through code.

Tommy points to a Vercel brand designer who had never coded before now being informally called a 'brand engineer,' and predicts every functional area will get its own '[blank] engineer' variant as AI tooling collapses the distance between deciding on a change and shipping it.

vibe-coding-adoption

The next stage beyond ad hoc AI tool use is a portable 'headless design' context layer - a briefcase of accumulated intent that travels between tools instead of being rebuilt each time.

Ramp built its own internal tool, Glass, specifically because general-purpose tools are good at general purpose while Glass is tuned to Ramp; Tommy argues this pattern (a company- or vertical-specific context layer, not a single general AI assistant) will repeat across design orgs and even households.

ai-context-layer

Evaluating AI models by single-prompt, one-shot comparisons ('GPT vs. Opus') is the wrong way to judge them; the real skill is learning to iteratively steer a model over a session.

Tommy compares it to test-driving a car: you have to correct the model repeatedly ('that's wrong, fix that') across several iterations to get a real feel for how it performs, and dismisses viral 'I compared three prompts' tweets as not measuring anything meaningful about design capability.

ai-fluency-hiring-signal

A weekly no-AI day preserves independent thinking and produces more authentic context to feed back into an AI workflow.

Tommy started forcing one full day per week without any AI tools two weeks before the recording; he found that without it, his own thinking became a crutch on AI output, and the no-AI day gives him genuinely self-authored context to hand back to the system afterward.

ai-context-layer

The best AI-native design teams succeed because leadership explicitly makes costly room for experimentation, not because of any particular tool.

At Vercel, a weekly Friday hackathon that started as a small informal thing grew into a company-wide event where sales, marketing, and other non-design functions present 'startup killer' prototypes; Tommy notes this is expensive in both token cost and throwaway work, and requires leaders to actively unblock time and spend.

design-org-restructuring

The three reasons designers say they aren't adopting AI tooling are outputs not being good enough yet, cost, and lack of company-provided time.

Tommy's survey found these three reasons roughly tied; he places the onus on leadership to make room, noting some designers already spend unpaid weekend time trying to keep pace so they don't look behind on Monday.

design-org-restructuring

Metalab mitigates the risk of company-wide AI workflow restructuring by isolating it to a small dedicated team first.

Its three-person 'Team Zero' picks a real client project, runs an AI-first workflow against its own success criteria, and reports results back, letting the rest of the org adopt incrementally instead of gambling on reinventing the whole 'factory floor' at once, the way Ramp has chosen to do company-wide.

design-org-restructuring

The strongest AI-fluency signal in hiring is demonstrated curiosity about workflows, not credentials or specific tool experience.

Rid says the number one red flag his hiring-manager network reports is a candidate showing no interest or curiosity in new AI processes; Tommy adds that publicly sharing experiments (e.g. on social media) has become the de facto way to prove you're exploring, since 'what are you showing that you're sharing' has become the discovery engine for evaluating designers.

ai-fluency-hiring-signal

Building small internal cross-functional tools is a practical career-differentiation strategy, not a form of self-replacement.

Tommy recommends designers find a cross-functional need nobody has time for (like marketing's blog images) and build a tool for it, citing Notion, Atlassian, and Stripe employees who got 'catapulted' internally after doing exactly this; he argues freed-up capacity gets redirected to higher-value product work, not eliminated.

internal-tool-building

Tommy predicts the next big conversation shift is toward an always-on AI collaborator embedded in the workflow, beyond today's context-briefcase stage.

He points to Dan Shipper's new writing tool as an early example of working alongside an AI that proactively flags relevant context ('you had a customer call last week that called out something interesting'), and expects onboarding and 'briefcase design' (what goes where, what's the taxonomy) to become a design problem in its own right over the next few quarters.

ai-context-layer

Media referenced

the 'brain' YC tweet - other - A viral tweet about a general-purpose AI 'design brain' that came up mid-conversation; Tommy has mixed opinions on the viability of a general-purpose context system at scale versus vertical-specific ones.
Dan Shipper's essays and tweets - article - Tommy fed a batch of bookmarked Dan Shipper articles and tweets into a single long Claude session that produced the nested CLAUDE.md context files he now uses to drive his design system.
Dan Shipper's new writing tool - other - Referenced near the end as a compelling example of working alongside an always-on AI collaborator that proactively surfaces relevant context, which Tommy sees as the next stage beyond today's 'briefcase of context.'
Amelia Wattenberger's blog posts - article - Cited as an example of a designer using freed-up time (from automating toil) to publish thoughtful writing combining generative images with interactive, philosophical essays about the future of interface work.

Companies

Vercel - One of the teams Tommy visited; runs a company-wide Friday hackathon where employees present 'startup killer' prototypes; Tommy says it's where he'd take a W-2 role tomorrow because of its builder-first culture.
Perplexity - One of the design teams Tommy visited on his tour of leading AI-native design orgs.
Ramp - Built an internal AI tool called Glass instead of adopting a general-purpose tool, because 'Co-work is good at general purpose, Glass is good at Ramp'; reportedly built roughly 1,500 internal vibe-coded tools, most of them throwaway.
Metalab - Runs a three-person unit called Team Zero that runs an AI-first workflow on real client projects (with client buy-in) and reports lessons back to the rest of the company, as a lower-risk way to restructure how the whole org works.
Figma - Still viewed as a core canvas tool even by the most AI-forward design teams Tommy has visited, as long as it keeps innovating on the agentic side.
Paper - A canvas tool Tommy is hearing about constantly in conversations despite it not appearing in the top 30 tools of his own survey data.
pencil.dev - A canvas tool Tommy flags as an interesting 'sneaky little winner' likely to keep improving.
Framer - Episode sponsor; Tommy/Rid discuss its new free AEO (AI-engine-optimization) scanner that scores how well AI systems like ChatGPT read a given website.
Jitter - Episode sponsor; a motion-design tool whose latest release lets users describe an effect in plain language and get tunable custom animation parameters.
InFlight - Rid's own product, a feedback tool for sharing prototypes and video walkthroughs, announced as newly in open beta during the episode.
Snowflake - Design engineer Will King, quoted secondhand, reflects on gaining systems-thinking capability while losing some of the free-form creative exploration process.
Conductor - The tool Tommy uses to run and monitor multiple parallel AI agent tasks/worktrees simultaneously, alerting him with a sound when each agent finishes.
Notion, Atlassian, Stripe - Named as companies where designers who built early internal AI-prototyping playgrounds got 'catapulted' internally in reputation and influence.
Shopify - Its internal-only 'artifact' tool went viral as one of the best-designed products of the year, illustrating how much visual polish internal tools can now reach.
Quantcast - Tommy's own earlier-career example of a scrappy, bootstrapped internal account-manager tool, contrasted with today's far more polished AI-built internal tools.
Oracle - Cited as an example of a large company where leadership is reportedly resistant to this AI-forward direction for design work.
Slack - Currently the best available interface for talking to AI agents, in Tommy's view, though he hopes Slack improves multi-thread agent conversation management.
Linear - Named as an example 'work lives here' surface-area tool that should sit alongside a canvas tool and a coding agent in a design org's core stack.
Google - Cited as launching a standardized design.md file, part of a broader industry push to standardize how design intent gets encoded for AI systems.

Techniques and frameworks

No-AI day - Tommy started forcing himself one day a week without any AI tools, to keep his independent thinking sharp and to generate authentic, self-authored context he can feed back into his AI setup afterward.
Driving instead of one-shotting - Tommy's framework for evaluating AI models: judge them by iteratively correcting and steering across a session ('that's wrong, fix,' repeated), not by comparing single one-shot outputs across models, which he says most public model comparisons wrongly do.
Headless design / context 'briefcase' - A portable, accumulating layer of context (visual rules, business logic, past stakeholder feedback) that should travel between AI tools rather than being rebuilt from scratch in each new tool; Ramp's Glass is offered as an early, company-specific instance of this.
Nested CLAUDE.md context files - Tommy built a hierarchy of markdown context files at different levels of his codebase in one dedicated multi-hour Claude session; he now never reads the files directly, only corrects outputs and lets Claude update the underlying rules.
Team Zero model - Metalab's approach: a small dedicated team runs a fully AI-driven workflow on one real client project, with its own success criteria, and reports findings back, letting the rest of the company adopt incrementally rather than reinventing every workflow at once.
Building internal cross-functional tools as a career strategy - Tommy's recommended tactic for designers wanting to stand out: find a cross-functional need (e.g., marketing's blog-post images) and build a small internal tool for it, which naturally makes you the internal subject-matter expert people start seeking out.

Summary

Tommy Geoco, who runs a state-of-prototyping survey and has spent the year visiting design teams at Vercel, Perplexity, Ramp, and Metalab, joins host Rid to synthesize what he's seeing across roughly 200 conversations with designers and design leaders. The episode opens with survey data: designers who vibe code heavily report higher satisfaction, but adoption skews toward design engineers, leads, and principals first, then non-designer roles, then managers, with junior and mid-level ICs adopting least - a pattern Tommy reads as either healthy trickle-down or evidence that some teams haven't made room for their whole staff yet. He also surfaces a widely misunderstood stat: 59% of designers have built their own internal tool, most of them narrow and disposable, which he argues is being unfairly lumped in with "AI slop."

A large portion of the conversation traces how both Tommy and Rid have restructured their own workflows around AI. Tommy describes building a hierarchy of nested markdown context files in a single multi-hour Claude session that he has never reopened - he only corrects outputs and tells Claude what never to do again, letting the system's implicit rules accumulate. Both hosts converge on a shared framework: judging AI models by single one-shot prompt comparisons is meaningless; the real skill is "driving" a model through iterative correction across a session, the way you'd test-drive a car. Tommy also describes starting a weekly no-AI day to keep his independent thinking sharp and to generate genuinely self-authored context to feed back into his system.

The episode's central thesis is what Tommy calls "headless design" - a portable context layer, or "briefcase," of accumulated intent, visual rules, and stakeholder feedback that should travel between tools instead of being rebuilt in each new one. He points to Ramp's internally built tool Glass as an early instance of this pattern, built specifically because general-purpose AI tools serve general purposes while a company-specific layer serves the company. From there, the conversation turns to how the best design orgs make deliberate, costly room for experimentation - Vercel's company-wide Friday hackathon and Metalab's dedicated "Team Zero" unit are offered as two different but complementary models for de-risking a full workflow overhaul, in contrast to Ramp's more blanket top-down mandate.

The back half shifts to career advice for designers navigating this shift. Tommy's clearest recommendation is to build small internal cross-functional tools - solving a marketing or sales team's problem that no one has time for - as a low-risk way to become the internal subject-matter expert people start seeking out, citing designers at Notion, Atlassian, and Stripe who were "catapulted" internally this way. On hiring, Rid shares that the single biggest red flag his network of hiring managers reports is a candidate showing no curiosity about new AI processes; both agree the strongest signal isn't tool credentials but visibly experimenting and sharing that experimentation publicly.

The episode closes with a prediction: over the next few quarters, the conversation will move past today's "briefcase of context" stage toward an always-on AI collaborator embedded directly in a designer's workflow, and the design of that context layer's onboarding and taxonomy will itself become a real design problem, one Tommy expects designers to be pulled into solving as much as engineers.

Notable Quotes

"This isn't AI slop. I'm making micro decisions that I didn't have a grid for a year ago and it's only because of the abilities that AI gives me." - Tommy Geoco

"I want the friction. There is a preserved level of friction that I'm finding is kind of important to me." - Tommy Geoco

"A red flag that I by far, the number one red flag that I see, is that a candidate has not demonstrated any interest or curiosity in new AI processes or tools." - Rid

"I think the future is bright, but it has to be built." - Tommy Geoco

"AI fluency is going to eventually just become like an ambient thing. I don't think we're going to highlight that language forever." - Tommy Geoco