https://mederic.me/blogMederic Burlet ยท Blog2026-10-05T00:00:00.000ZMederic Burlethttps://mederic.meNeurowirehttps://mederic.me/blog/oqf-freezing-a-format-at-v0-1Why Open Quest Format froze its file format at v0.1 while the packages keep changing2026-10-05T00:00:00.000Z2026-10-05T00:00:00.000ZMederic Burlethttps://mederic.meOpen Quest Format has two version numbers. The file format was frozen at OQF1 in the first release and can only grow by appending columns, while the TypeScript packages stay 0.x and are still allowed to break their API. This post explains why the two are kept apart and how the append-only rule works.https://mederic.me/blog/oqf-dialogue-as-a-siblingWhy Open Quest Format keeps dialogue separate from quests2026-10-04T00:00:00.000Z2026-10-04T00:00:00.000ZMederic Burlethttps://mederic.meIn Open Quest Format, conversations live in their own model. A quest step points at a dialogue by reference and only receives the flags and events the dialogue sets, which is what makes it possible to plug in Ink and Yarn Spinner.https://mederic.me/blog/rust-native-module-four-targetsShipping a Rust native module to four targets from one CI2026-10-03T00:00:00.000Z2026-10-03T00:00:00.000ZMederic Burlethttps://mederic.meHow cozy-steamworks turns one Rust crate into four prebuilt .node binaries, packs them into a tarball on GitHub Releases, and why a plain git dependency on the repo cannot work.https://mederic.me/blog/oqf-what-every-quest-has-in-commonWhat Skyrim and Baldur's Gate 3 quests have in common2026-10-02T00:00:00.000Z2026-10-02T00:00:00.000ZMederic Burlethttps://mederic.meA walkthrough of the Open Quest Format model: how steps form a flat graph, how objectives count events, how rewards and endings are emitted as events and flags, and why so much is left to the engine on purpose.https://mederic.me/blog/steam-leaderboards-global-stats-gotchasAdding leaderboards and global stats to steamworks.js: what Steam doesn't tell you2026-10-01T00:00:00.000Z2026-10-01T00:00:00.000ZMederic Burlethttps://mederic.meNotes from adding leaderboards, aggregated global stats and screenshots to my steamworks.js fork, covering handles that only last one Steam session, throttled uploads that resolve as if nothing went wrong, and global totals that run about a day behind.https://mederic.me/blog/oqf-compact-formatMeasuring the OQF compact format: 677 tokens against 1,200 for JSON2026-09-30T00:00:00.000Z2026-09-30T00:00:00.000ZMederic Burlethttps://mederic.meI measured Open Quest Format's compact serialization against its own JSON output on the reference quest to see where the bytes and tokens go, and what the line format costs in return.https://mederic.me/blog/steam-callbacks-cant-wait-for-javascriptDeciding Steam session requests in Rust, before JavaScript can answer2026-09-29T00:00:00.000Z2026-09-29T00:00:00.000ZMederic Burlethttps://mederic.meISteamNetworkingMessages expects a session to be accepted or rejected before its callback returns, and JavaScript can't run in time. So my steamworks.js fork decides from a Rust-side allow list that JavaScript fills in ahead of time.https://mederic.me/blog/oqf-provably-completable-questsProving a quest is completable before anyone plays it2026-09-28T00:00:00.000Z2026-09-28T00:00:00.000ZMederic Burlethttps://mederic.meOpen Quest Format keeps its condition language to eleven operators and no scripts, so that the validator can find unreachable steps, dead ends and quests with no reachable ending. This post explains how the search works and what that limit costs.https://mederic.me/blog/forking-steamworks-jsForking steamworks.js for Cozy Coast's multiplayer hub2026-09-27T00:00:00.000Z2026-09-27T00:00:00.000ZMederic Burlethttps://mederic.meUpstream steamworks.js 0.4.0 only offered legacy P2P, an unfiltered lobby list and no friends API, which was not enough for Cozy Coast's multiplayer. This covers what the fork added, what it removed, and what it costs to maintain a fork for a single game.https://mederic.me/blog/oqf-no-engine-owns-your-questsWhy I wrote an open quest spec instead of a plugin2026-09-26T00:00:00.000Z2026-09-26T00:00:00.000ZMederic Burlethttps://mederic.meOpen Quest Format 0.1.0 is an Apache-licensed, versioned quest spec with a reference parser, validator and runtime, so quest content can outlive the engine or plugin you picked first.https://mederic.me/blog/neurowire-sync-federationThe laptop that was closed all day2026-09-03T00:00:00.000Z2026-09-03T00:00:00.000ZMederic Burlethttps://mederic.meNWF Sync is a four-route read-only protocol for pulling feed journals between machines, built for the case direct fetching cannot solve: items that already scrolled off the front page while you were away.https://mederic.me/blog/neurowire-tail-streamingTail: following a feed instead of re-fetching it2026-09-02T10:30:00.000Z2026-09-02T10:30:00.000ZMederic Burlethttps://mederic.meNeurowire 0.8.0 replaces every polling loop in the codebase with one async generator, so the CLI, watch mode, and the server-sent-events endpoint cannot drift on cadence, dedupe, or politeness.https://mederic.me/blog/neurowire-tap-wizardA tap wizard with zero LLM calls2026-09-01T12:00:00.000Z2026-09-01T12:00:00.000ZMederic Burlethttps://mederic.meNeurowire 0.8.0 generates CSS-selector taps by ranking candidates from page structure and gating every one behind eight deterministic checks, with no model and no API key anywhere in the loop.https://mederic.me/blog/press-kit-as-codeGenerating a press kit from the running game2026-08-26T10:30:00.000Z2026-08-26T10:30:00.000ZMederic Burlethttps://mederic.meTwelve social cards built from HTML and CSS templates, filled with real screenshots captured from the running game, and rebuilt with one command instead of a design tool.https://mederic.me/blog/one-changelog-three-surfacesOne changelog file, three places it shows up2026-08-25T00:00:00.000Z2026-08-25T00:00:00.000ZMederic Burlethttps://mederic.meA single typed changelog file feeds an in-game badge, a Settings tab and a public website page, so release notes cannot drift between the product and its marketing.https://mederic.me/blog/cozy-coast-tally-no-accountsA public tally with no accounts and no IDs2026-08-23T10:30:00.000Z2026-08-23T10:30:00.000ZMederic Burlethttps://mederic.meCozy Coast ships analytics that count events instead of people: opt in, no account, no device ID, and a public page as the whole reporting layer. Here is what that costs.https://mederic.me/blog/cozy-coast-beta-numbersWhat the beta tally told me about my own game2026-08-22T00:00:00.000Z2026-08-22T00:00:00.000ZMederic Burlethttps://mederic.meThe Cozy Coast beta counted 160,966 catches and 50,418 seagull greetings. The action that pays nothing was the most repeated one, and the economy changed because of it.https://mederic.me/blog/cozy-coast-desktop-widget-electronBuilding a game that behaves like a desktop widget2026-08-21T10:30:00.000Z2026-08-21T10:30:00.000ZMederic Burlethttps://mederic.meCozy Coast is an Electron and Phaser game built as a borderless desktop widget first, which changes the window model, the render budget and the packaging pipeline.https://mederic.me/blog/cozy-coast-multiplayer-no-serversCozy Coast multiplayer runs on zero servers2026-08-21T00:00:00.000Z2026-08-21T00:00:00.000ZMederic Burlethttps://mederic.meCozy Coast shipped a multiplayer hub with no backend at all. Steam handles lobby discovery, connection brokering and packet relay, and the host relays player state.https://mederic.me/blog/ghostwriter-getting-startedGetting started with Ghostwriter, step by step2026-07-20T10:30:00.000Z2026-07-20T10:30:00.000ZMederic Burlethttps://mederic.meA plain walkthrough of installing and using the free Ghostwriter tool: where the files go, how to ask for a rewrite, what the clean-up step does, and how much of a difference it actually made in testing.https://mederic.me/blog/ghostwriter-spot-ai-writingHow to spot AI writing before you hit send2026-07-19T10:30:00.000Z2026-07-19T10:30:00.000ZMederic Burlethttps://mederic.meA quick, non-technical checklist for catching the giveaway signs of AI-written text in your own drafts: the vocabulary, the lists of three, the flat rhythm, the long dash, and the too-neat ending.https://mederic.me/blog/langdrift-translation-driftLangdrift: scoring how differently a translation lands2026-07-19T00:00:00.000Z2026-07-19T00:00:00.000ZMederic Burlethttps://mederic.meLangdrift compares audio of an original and its translation and returns a drift score: how far apart the two land cognitively, catching accurate but off translations.https://mederic.me/blog/ghostwriter-why-humanizing-failsWhy 'make it sound human' never quite works2026-07-18T10:30:00.000Z2026-07-18T10:30:00.000ZMederic Burlethttps://mederic.meAI writing gives itself away on three levels: the characters, the words, and the shape of the whole piece. The deepest one is decided while the text is being written, which is why a clean-up pass at the end cannot fully fix it.https://mederic.me/blog/verso-json-presentationsVerso: one JSON source, decks branched per audience2026-07-18T00:00:00.000Z2026-07-18T00:00:00.000ZMederic Burlethttps://mederic.meVerso defines presentations as JSON instead of slide documents, so one source can branch per audience and export to PDF, HTML, or PNG cleanly.https://mederic.me/blog/ghostwriter-human-writingAI writing has a tell. I built a free tool that removes it.2026-07-17T10:30:00.000Z2026-07-17T10:30:00.000ZMederic Burlethttps://mederic.meGhostwriter is a free, open-source tool that makes AI-written text read like a person wrote it, by removing the hidden characters, the giveaway words, and the report-shaped structure that flag it as machine-made.https://mederic.me/blog/petrify-on-devicePetrify: removing people from footage, entirely on-device2026-07-17T00:00:00.000Z2026-07-17T00:00:00.000ZMederic Burlethttps://mederic.mePetrify detects and removes people from fixed-camera footage while keeping the live timestamp ticking, running fully on your own machine with no cloud upload.https://mederic.me/blog/scope-beats-autonomyScope beats autonomy: a contrarian take on AI agents2026-07-16T00:00:00.000Z2026-07-16T00:00:00.000ZMederic Burlethttps://mederic.meOpen-ended autonomous agents make great demos and poor production systems. Tightly scoped, observable, boring agents are what actually survive real workloads.https://mederic.me/blog/multi-agent-reasoningMulti-agent workflows a single engineer can reason about2026-07-15T00:00:00.000Z2026-07-15T00:00:00.000ZMederic Burlethttps://mederic.meA focused multi-agent system one engineer can reason about beats a sprawling one nobody trusts. Scope each agent, define tools explicitly, and observe everything.https://mederic.me/blog/measuring-ai-productivityAn honest look at AI productivity claims, and how to measure 2-3x2026-07-14T00:00:00.000Z2026-07-14T00:00:00.000ZMederic Burlethttps://mederic.meA 2-3x productivity gain is real when you measure build success, cycle time, and effort reduction, not vanity metrics. Here is how to tell the difference.https://mederic.me/blog/evals-for-delivery-teamsEvals for delivery teams: trusting AI output without re-checking every line2026-07-13T00:00:00.000Z2026-07-13T00:00:00.000ZMederic Burlethttps://mederic.meTrustworthy AI output comes from evals built into the real pipeline, not a benchmark: evaluate what matters, gate at the right point, and keep review habits that scale.https://mederic.me/blog/no-code-ai-workflowBuild a tiny no-code AI workflow that saves you an hour a week2026-07-12T10:30:00.000Z2026-07-12T10:30:00.000ZMederic Burlethttps://mederic.meChain a trigger, an AI step, and an action in a no-code tool like Zapier, Make, or n8n to automate one small repetitive task end to end, no programming needed.https://mederic.me/blog/make-yourself-unnecessaryMake yourself unnecessary: what good advisory leaves behind2026-07-12T00:00:00.000Z2026-07-12T00:00:00.000ZMederic Burlethttps://mederic.meGood advisory leaves a team stronger, not dependent. Capability and judgement should stay with the team, through co-delivery, explicit reasoning, and grown champions.https://mederic.me/blog/make-a-song-with-sunoMake a full song with AI in ten minutes using Suno2026-07-11T10:30:00.000Z2026-07-11T10:30:00.000ZMederic Burlethttps://mederic.meDescribe the song you want, pick a style, generate a couple of takes, refine the parts you do not like, then export. Suno makes a full track in minutes.https://mederic.me/blog/predicting-ai-adoptionWhat predicts whether an AI cohort actually sticks2026-07-11T00:00:00.000Z2026-07-11T00:00:00.000ZMederic Burlethttps://mederic.meAcross 700+ practitioners, durable AI adoption tracks three predictors: training on their own codebase, role-specific tracks, and internal champions seeded in the team.https://mederic.me/blog/nano-banana-explainedNano Banana, explained: what Google's image model is good at2026-07-10T10:30:00.000Z2026-07-10T10:30:00.000ZMederic Burlethttps://mederic.meNano Banana is the nickname for Google's Gemini image model, great at quick edits and keeping a subject consistent across pictures, with a few known limits.https://mederic.me/blog/sia-copilot-anatomySingapore Airlines: a website feature in five weeks, down from nine2026-07-10T00:00:00.000Z2026-07-10T00:00:00.000ZMederic Burlethttps://mederic.meEmbedding GitHub Copilot across Singapore Airlines' delivery lifecycle cut a feature from nine weeks to five, with 95% of the work AI-generated and effort down 60%.https://mederic.me/blog/local-vs-cloud-aiLocal vs cloud AI: when to run models on your own machine2026-07-09T10:30:00.000Z2026-07-09T10:30:00.000ZMederic Burlethttps://mederic.meRun AI locally for privacy, no per-use cost, and offline use if you have decent hardware, or use cloud AI when you want the most capable models with zero setup.https://mederic.me/blog/three-training-tracksAI training needs a track per role, not one shared workshop2026-07-09T00:00:00.000Z2026-07-09T00:00:00.000ZMederic Burlethttps://mederic.meEngineers, leads, POs, QA, designers, and managers use AI for different work, so one shared workshop gives everyone half-relevant material. Role-specific tracks outperform it.https://mederic.me/blog/replicate-run-any-modelReplicate: run almost any AI model without the setup2026-07-08T10:30:00.000Z2026-07-08T10:30:00.000ZMederic Burlethttps://mederic.meReplicate lets you run open AI models for images, audio, video, and text through a simple interface, with no GPU or environment to set up on your side.https://mederic.me/blog/demo-to-production-gapThe demo-to-production gap is an organisational problem2026-07-08T00:00:00.000Z2026-07-08T00:00:00.000ZMederic Burlethttps://mederic.meA better model does not close the demo-to-production gap. The skills that win a pilot do not generalise, so architecture, guardrails, and behaviour change close it.https://mederic.me/blog/voice-cloning-elevenlabsCloning a voice with ElevenLabs, responsibly2026-07-07T10:30:00.000Z2026-07-07T10:30:00.000ZMederic Burlethttps://mederic.meYou can clone a voice with ElevenLabs in minutes, but only do it with a voice you own or have clear permission to use, never someone else's.https://mederic.me/blog/adoption-behaviour-changeAdoption is a behaviour change, not a tooling rollout2026-07-07T00:00:00.000Z2026-07-07T00:00:00.000ZMederic Burlethttps://mederic.meTools do not change behaviour, practice does. A launch plus a generic workshop spikes then fades, while role-specific practice and champions make adoption durable.https://mederic.me/blog/spotting-ai-slopHow to spot AI slop and keep your own output good2026-07-06T10:30:00.000Z2026-07-06T10:30:00.000ZMederic Burlethttps://mederic.meAI slop is low-effort, generic text with vague filler, endless hedging, and no sources, and you avoid it by adding specifics, your own voice, and real editing.https://mederic.me/blog/prompting-is-editingPrompting is editing, not incantation2026-07-06T10:30:00.000Z2026-07-06T10:30:00.000ZMederic Burlethttps://mederic.meThe real skill with AI is not finding magic words, it is giving feedback. Draft, say what is wrong, refine, repeat until the answer is right.https://mederic.me/blog/modern-prompting-no-roleplayStop role-playing your prompts: what works on modern AI2026-07-06T10:30:00.000Z2026-07-06T10:30:00.000ZMederic Burlethttps://mederic.meYou no longer need to tell modern AI it is a world-class expert or beg it to think step by step. Give it a clear goal, real context, and the format you want.https://mederic.me/blog/anatomy-of-a-good-promptThe anatomy of a good prompt, and three myths to drop2026-07-06T10:30:00.000Z2026-07-06T10:30:00.000ZMederic Burlethttps://mederic.meA good prompt is just five plain parts: a goal, real context, the output format, a few constraints, and an example or two. No magic words required.https://mederic.me/blog/ai-image-gen-replicate-nano-bananaAI image generation, made simple: Replicate and Nano Banana2026-07-06T01:49:17.000Z2026-07-06T01:49:17.000ZMederic Burlethttps://mederic.meWant AI images without a powerful GPU? Use Nano Banana for quick photo edits, Replicate to try many models in your browser, and a LoRA for a repeatable style.https://mederic.me/blog/neurowire-taps-selectorsTaps: turning feed-less websites into real feeds2026-07-06T00:00:00.000Z2026-07-06T00:00:00.000ZMederic Burlethttps://mederic.meNeurowire taps are per-host CSS-selector recipes that turn a plain HTML listing page into a real feed, so a site with no RSS becomes a first-class source.https://mederic.me/blog/never-paste-into-chatbotWhat you should never paste into a chatbot2026-07-05T10:30:00.000Z2026-07-05T10:30:00.000ZMederic Burlethttps://mederic.meAssume anything you paste into a chatbot may be stored or used for training, so never share passwords, secrets, other people's data, or confidential work.https://mederic.me/blog/dogfooding-neurowire-feedsDogfooding @neurowire/core to power this blog's feeds2026-07-05T00:00:00.000Z2026-07-05T00:00:00.000ZMederic Burlethttps://mederic.meThis site publishes its own Atom, JSON Feed, and NWF feeds from one canonical Neurowire model, because dogfooding your own library is the most honest test it gets.https://mederic.me/blog/understanding-lorasUnderstanding LoRAs without the jargon2026-07-04T10:30:00.000Z2026-07-04T10:30:00.000ZMederic Burlethttps://mederic.meA LoRA is a small add-on file that teaches an image model one style, character or subject without retraining the whole thing, and you can stack a few.https://mederic.me/blog/nocturne-design-systemNocturne: an open-source Cyberpunk 2077 design system2026-07-04T06:21:07.000Z2026-07-04T06:21:07.000ZMederic Burlethttps://mederic.meNocturne is an open-source design system inspired by Cyberpunk 2077 and Edgerunners: near-black surfaces, one hot neon accent, clipped corners, and machine-voice type, shipped as design tokens, framework-agnostic CSS, and React components.https://mederic.me/blog/building-neurowireBuilding Neurowire: one canonical model, six feed formats2026-07-04T00:00:00.000Z2026-07-04T00:00:00.000ZMederic Burlethttps://mederic.meNeurowire treats feeds as a data-modelling problem: one canonical model that every parser produces and every serializer reads, so adding a format is a single serializer.https://mederic.me/blog/everyday-ai-toolkitThe everyday AI toolkit: Claude, Suno, ElevenLabs and friends2026-07-03T10:30:00.000Z2026-07-03T10:30:00.000ZMederic Burlethttps://mederic.meMatch the job to the tool: Claude for writing and thinking, Suno for music, ElevenLabs for voice, and Replicate for running image models like Flux.https://mederic.me/blog/open-web-feedsThe open web forgot about feeds, and that is worth fixing2026-07-03T00:00:00.000Z2026-07-03T00:00:00.000ZMederic Burlethttps://mederic.meFeeds let you publish once and let anything read it. As they eroded, following became an account on someone else's platform. Owning what you read is independence.https://mederic.me/blog/ai-for-your-day-10-winsAI for your actual day: ten small wins most people miss2026-07-02T11:30:00.000Z2026-07-02T11:30:00.000ZMederic Burlethttps://mederic.meAI is most useful for small daily tasks like drafting replies, summarising documents, planning trips, and tidying spreadsheets, not just for big flashy projects.https://mederic.me/blog/nwf-format-designDesigning NWF: a feed format 62% smaller than JSON Feed2026-07-02T10:30:00.000Z2026-07-02T10:30:00.000ZMederic Burlethttps://mederic.meNWF is Neurowire's native feed format, roughly 62% smaller than the equivalent JSON Feed yet fully round-trippable, using interning, relative links, and date deltas.https://mederic.me/blog/fact-check-ai-answerHow to fact-check an AI answer in 30 seconds2026-07-01T13:00:00.000Z2026-07-01T13:00:00.000ZMederic Burlethttps://mederic.meOpen the cited source and check it actually says what the AI claims, then cross-check one independent source, paying closest attention to numbers, dates, and names.https://mederic.me/blog/unremarkable-workflowWhy I optimise for the unremarkable workflow, not the wow demo2026-07-01T10:00:00.000Z2026-07-01T10:00:00.000ZMederic Burlethttps://mederic.meA hundred engineers using AI by default beats one jaw-dropping demo. Production compounds on the thousandth unremarkable run, so measure throughput, not wow.https://mederic.me/blog/neurowire-taps-packNeurowire Taps Pack: 271 curated sources, 24 themes2026-06-29T00:00:00.000Z2026-06-29T00:00:00.000ZMederic Burlethttps://mederic.meThe Taps Pack is a ready-made library of 271 vetted sources for Neurowire, grouped into 24 themes from Frontier AI Labs to Food. It's the curation layer ahead of the upcoming Neurowire SaaS.https://mederic.me/blog/neurowire-docs-launchNeurowire: an open feed engine, now with full docs2026-06-28T00:00:00.000Z2026-06-28T00:00:00.000ZMederic Burlethttps://mederic.meNeurowire turns any source, even feed-less websites, into one canonical feed you can serialize to Atom, JSON Feed, Markdown, RSS, or its own compact NWF format. The full documentation site is now live.https://mederic.me/blog/why-ai-pilots-stallWhy AI pilots stall before production2026-06-25T00:00:00.000Z2026-06-20T00:00:00.000ZMederic Burlethttps://mederic.meAI pilots stall because they optimise for a demo, not for adoption. Production requires architecture, guardrails, and a change in how teams work, not a better model.https://mederic.me/blog/700-engineers-on-adoptionWhat 700 trained engineers taught me about adoption2026-05-12T00:00:00.000Z2026-05-12T00:00:00.000ZMederic Burlethttps://mederic.meAfter training 700+ practitioners, the pattern is clear: adoption is a behaviour change, not a tooling rollout. Role-specific practice and internal champions are what make it stick.