Three days, nineteen talks, and a lot to think about.
I spent the last three days at Develop:Brighton 2026, sitting in on talks about Steam marketing data, reinforcement learning, UI design, and one gloriously unfiltered argument against conflict-driven storytelling. This isn’t a session-by-session recap. It’s what kept resurfacing once I put all nineteen talks side by side: some threads were obvious from the programme (funding, discoverability, AI), others only showed up once unconnected speakers started saying the same thing without appearing to have compared notes.
The industry isn’t pretending things are fine
Nobody at this conference tried to talk around the state of things, which was a relief. The clearest evidence came from the UK Games Industry Workforce Demographics survey, presented on behalf of Ukie’s Raise the Game programme by Professor Mark Taylor, its author. Of the 1,610 people who completed it, around 22% had experienced an involuntary job loss (redundancy, a fixed-term contract that wasn’t renewed, or a studio closing) in the last three years. Roughly 80% said they were proud to tell people where they work, and 70% would recommend their specific employer. Only 38% would recommend the UK games industry as a whole. That gap, between loving your studio and being wary of the industry it sits inside, is the most honest snapshot of morale I saw all week.
Several other talks filled in why. Alyx Jones, founder of Silver Script Games, explained she’d been made redundant from a post-production audio role (she’d worked on Elden Ring’s dialogue) three years before shipping her own first game. Murray Pannell’s talk on building an indie studio from nothing began the same way: six people from a studio whose AAA shooter had run out of funding, deciding to build something themselves rather than “doom-scroll LinkedIn.” Amir Didarzadeh argued, more analytically, that a lot of this is self-inflicted: AAA studios routinely mistake a budget-driven spreadsheet target for an actual market forecast, then spend years discovering the gap. Even Adele Cutting’s talk on the history of voice performance and motion capture ended on freelancers still routinely waiting 60 to 90 days to get paid.
None of this is news if you’ve read the trade press over the last couple of years. But hearing it from unconnected speakers across three days, in rooms full of people nodding along, lands differently than a headline does.
Funding a game now means stitching together five different things at once
Alyx Jones’s talk was the most granular account of what paying for a game looks like in 2026: a patchwork. Funding for The Quiet Things came from an Innovate UK-style prototype grant, a Kickstarter campaign (“the golden days of Kickstarter are over,” she said), a women-in-innovation grant, an Xbox creator fund, the UK’s Video Games Expenditure Credit, and paid academic collaboration on the side: £150,000 in total. Publishers were avoided, because the deals are getting worse and publishers now want proven retention data over a good story. Her advice was blunt: register a company early, build a vertical slice before pitching because people respond to what they can see, and put your Steam build up two months before launch, since certification delays are unpredictable and there’s no one to call when they happen.
Big Dumb Games, Murray Pannell’s studio, is the more radical answer: no publisher, no investor, no crowdfunding to start with, just a “craft funding” model where contributors bank hours as points against a future profit share on a spreadsheet ledger. It runs as a cooperative rather than a company: six co-founders who steer rather than own, with a founders’ agreement they’d already had to rewrite once contribution levels diverged from the original equal split. Whether that survives contact with an actual publisher deal is an open question.
Getting found: relationships still beat spend
Three talks tackled discoverability from different angles and mostly arrived at the same conclusion. Helen Carmichael of Grey Alien Games gave the most data-rich case study, built around Forbidden Solitaire, a retro FMV horror card game made with Night Signal Entertainment (of Home Safety Hotline fame): a year of marketing before the demo even existed, 35 to 36 Steam events entered over that period, a teaser trailer that generated 7,000 wishlists in two days when all they had was a design document, and a demo launch that hit number one on Steam’s New & Trending Demos off roughly 10,000 downloads in under a week. Notably, only 3.6% of the traffic to the store page during that spike came from Steam’s trending page itself; most came from relationships they’d built with a small, curated list of horror streamers over the preceding year. Her summary: “connections matter.”
Gary Burchell’s talk on Steam events backed this up with harder numbers from six years running the Guildford Games Festival: over ten and a half million cumulative page visits, with the smallest, least-known games in the lineup seeing proportionally the biggest gains. One small title picked up a 30% wishlist increase against a flagship title’s 1% in the same event; a solo developer credited the festival with visibility she couldn’t get alone, later funding her next game from it and growing into a four-person studio. Events aren’t a silver bullet, though: the three biggest wishlist spikes on SENTRY, the follow-up to his Abandon Ship, came from a Popular Upcoming placement, a Steam Next Fest demo slot, and a viral Reddit post, not a hosted event at all. Events did roughly a third of the overall work, but only alongside everything else.
Guy Richards, Global Director of Xbox’s ID@Xbox programme, reframed discoverability as “lots of small things” rather than one big lever: plan your announcement timing before you announce (going public opens doors like wishlisting while closing others, like eligibility for a platform showcase reveal), talk to platforms earlier than feels necessary, and treat your store page as a sales tool you keep editing rather than a form you fill in once. The most memorable tip was the oldest one in the room: screenshot your box art next to a dozen competitors’ and see if it still stops a scroll.
And then there was a talk arguing all of this is about to gain a new layer entirely. Ravi Vijh, managing director of the games comms agency Bastion, made the case that players are starting to ask AI chatbots what to play rather than searching for it, and that large language models lean on earned media and consistent, public content (developer diaries, forum activity, the same message repeated everywhere) over a studio’s own marketing copy. A live demo asking ChatGPT for a co-op game returned not a big-budget title but a long-running early access game with years of consistent community engagement behind it. It’s an early, speculative thesis, he admitted, but a genuinely different discoverability problem to the ones the other talks were solving.
The craft conversations: mostly about asking why, not copying what
The strongest common thread in the design talks wasn’t a shared technique, it was a shared discipline: understand why something works before you borrow it, rather than lifting the surface mechanic and hoping the feeling comes with it.
Caspar Gray, who spent a decade signing games at Eidos and Square Enix and another as a publisher-pitching agent before landing at Green Man Gaming, put it as directly as anyone all week: ask why you’re making something, not how, and when you pitch it, “convince, don’t describe.” The best illustration all week was Rocket League’s history: Psyonix’s soccer mode started as a joke feature buried inside a much bigger, more baroque game (originally pitched as Supersonic Acrobatic Rocket-Powered Battle-Cars), and the follow-up’s entire success came down to stripping everything else away and renaming it something a search engine and a first-time player could both parse instantly. Annie Grudeva, lead designer at Radical Forge, built the same idea into a formal “STEAL” framework (syntax, tension, emotion, alignment, lock), illustrated with Hitman versus IO Interactive’s new James Bond game, 007 First Light: both stealth games from the same studio, deliberately tuned to produce opposite player behaviour (disguise in one, charisma in the other), because copying the mechanic without copying the intent would only produce a worse Bond game.
Lydia Cockerham, principal narrative designer at Creative Assembly who previously worked on Dead Island 2, offered the best answer I heard to the problem of AAA narrative design under constant, brutal scope change. Her “Dream Mode” framework treats story elements as swappable archetypes rather than fixed lore, so that when a studio suddenly cuts your commander character, boss count, and sci-fi setting altogether, you can rebuild around what’s left instead of starting over. Tristan Jacobs, a Glowmade designer with a theatre background, offered a companion framework: the “machete” (whoever makes the first decisive cut into an idea) and the “broom” (whoever aligns everything that follows), illustrated with a detail from Hellboy: Web of Weird, where the team kept a Mike Mignola comic panel pinned to the director’s desk and periodically asked “are we still doing this” against it.
Agnes Kasilovska and Sam Thompson, of Dead Nice Studio, made the case that UI and branding are the same discipline wearing different clothes, and that treating them as sequential jobs (make the game, then skin it, then market it) is why so many games’ marketing promises don’t survive contact with their menus. Ghost of Tsushima‘s wind mechanic, doubling as both navigation UI and the visual signature of its key art, was their clearest example. Shi-tao Fan, who moved from narrative encounter work on Avatar: Frontiers of Pandora to Cloud Imperium Games, gave the most detailed account of what that discipline costs at AAA scale: a target of 500 vignette-style encounters became roughly 300 after a year’s delay and a budget cut, and the team survived it by choosing “staging” (reusable, remixable tableaux) over deep branching narrative, having decided early they couldn’t afford the Fallout- or Red Dead-style approach they’d first benchmarked against.
And the talk that argued hardest against received wisdom, delivered with more energy than anything else on the schedule, made the case that conflict isn’t the universal engine of drama it’s taught as. Its speaker, Christopher Morrison, a writer-director working across theatre, XR, and games, pointed to Kishotenketsu, the Japanese four-beat structure of introduction, development, twist, and conclusion rather than escalating opposition, and to the commercial success of cozy games and romance fiction (citing Spiritfarer‘s reported $25 to 28 million from a small Canadian studio, among other figures) as evidence that curiosity, cooperation, and unresolved questions can carry a story just as well as a villain can. Suggestions included subverting win-loss binaries rather than resetting on failure, and giving non-player systems (weather, ecology, NPC networks) their own agency, so the player becomes a participant rather than the sole protagonist. Whatever you make of the framing, the underlying point, that growing a character doesn’t have to mean hurting them, is worth sitting with.
AI: not one conversation, at least three
AI came up constantly, and what struck me was how little consensus there was, even within the same building.
Rob Davis’s talk, from his studio Wild Loop Games, was the most technically serious. He spent two and a half months, part-time, alone, training reinforcement-learning models to play card games, getting one to a 97.9% win rate against a mid-tier heuristic opponent in Sushi Go, 39.2% for his Uno model, 64.6% on Whot, and 57.6% on Sea Salt & Paper. He was upfront that AI-generated code sprawls in ways that quietly corrupt a project (duplicate variables silently showing different scores on different screens was his example), and that fixing it meant a far more disciplined, test-first, human-reviewed pipeline that roughly doubled his development time. He called that a good trade, not a failure, because it left him with a documented, portable project that survives a model being deprecated or a chat history auto-deleting, as happened to him once. He also opened by naming the legitimate concerns around training data, energy and water use, and employment impact as worth taking seriously rather than waving away.
Big Dumb Games sits at the other end: a founding studio value, stated plainly by Murray Pannell, to build with human craft and not touch AI at all, partly on principle and partly because their artists find it difficult to work alongside.
And the recommendation-era talk treated AI as neither a development tool nor a threat, but as a new discovery channel that’s going to matter whether or not a given studio uses it in production at all. Three very different relationships to the same technology, all represented at one conference, none of them obviously wrong. I don’t think the indie scene has settled this, and I’m not sure it should settle it quickly.
Who gets to make games, and what happens to them
The workforce survey had sharper findings than the headline numbers above. Neurodivergent conditions are reported at far higher rates than the general population (27% reported a condition affecting concentration, such as ADHD, and 16% autism), partly because people now have the language to recognise it in themselves, a positive shift even where the numbers are sobering. Representation thins out sharply with seniority: senior leadership is around 72 to 73% men, and LGBTQIA+ representation, well above the general population overall, skews markedly younger and more junior. Job losses weren’t evenly distributed either: people with a nationality from outside the UK were more likely to have lost a job than UK nationals, and women and non-binary colleagues more likely than men.
The ADHD roundtable, moderated by Rory Martin, put a human face on that data from the inside: attendees swapping coping strategies (external memory systems, timed work sprints, deliberately hyperfixating on the QA edge cases nobody else wants to touch), alongside a sharper point repeated throughout, that burnout in neurodivergent staff usually traces back to managers with no training, not the condition itself. Cinzia Musio, presenting Ukie’s response, set out the Raise the Game coalition’s plan to close that gap: manager training, transparent progression tracking, and restarting community meetups for marginalised groups.
The talk that sat least comfortably in a straightforward “industry trends” bucket, and the one I keep thinking about most, came from Stanley W. Baxton, a narrative designer and University of Salford PhD researcher who makes deeply personal, non-commercial interactive fiction for a narrow online community. He was startled when LATEX, LEATHER, LIPSTICK, LOVE, LUST, his most recent project, connected far beyond that intended audience, well enough to be selected for BAFTA Breakthrough, something he’d half-expected to be laughed out of the room for submitting. His broader argument, that the industry quietly equates “video game” with “commercial industry product” and treats anything outside that as a hobbyist’s stepping stone rather than a finished statement, is harder to summarise fairly than the marketing talks were, but it’s a useful corrective at a conference that is, inevitably, mostly about how to sell things.
Why we do this
It would be easy to read all of that back and conclude the industry is having a hard time, because it is. Layoffs, funding freezes, studio closures, an industry only 38% of its own workforce would currently recommend: none of that is exaggerated, and none of it went unmentioned by the people living through it.
But that’s not what I actually took away from three days at Develop. What I took away was a room full of people who’d been made redundant and started their own studios anyway, who’d run a Steam events page for six years to give someone else’s first solo game a leg up it couldn’t have got alone, who took their trained AI back to a kids’ board game club because that’s where the idea started, who kept a comic panel taped to a director’s desk to remind themselves what they were actually trying to make. Behind every brand, every logo, and every game on that expo floor, there was just a person, or a small handful of them, who cared enough to turn up and talk about why they do this and how to do it better. That’s what an event like this is actually for: not just the panels and the pitch decks, but a place to learn, build a network, and swap ideas with people who understand exactly what you’re going through, because a lot of them are too.
I left Brighton positive about where all this is heading, and more determined than ever to keep making games. That’s the whole point of turning up.