Adobe Has a Big Problem  🔗 **Source:** [Adobe Has a Big Problem](https://www.youtube.com/watch?v=SL8LD9oxVlY) > [!note]+ Document Notes > tags: [clean-room-porting, adobe-alternatives, vibe-coding, design-software, mcp-access, software-moats] > author: Syntax > source: Syntax (YouTube) > reference: https://www.youtube.com/watch?v=SL8LD9oxVlY > date created: 2026-10-14 08:46 > > > [!info] Generated analysis — not the author's words. Sections below are machine-written > > from the captured source. Only text inside `> [!quote]` blocks is verbatim from the author. > > Do not quote or attribute any other line to the named author. > > # 🏷️ Title: Adobe has a big problem > <!-- deep-read:sections start --> ## 🧭 Table of contents - [📝 Summary](#-summary) - [📖 Main content](#-main-content) - [🗒️ Notes](#️-notes) - [📚 References](#-references) - [🗣️ Paraphrase](#️-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning styles assessment](#-learning-styles-assessment) - [📊 Accessibility assessment](#-accessibility-assessment) - [🌍 DEIB assessment](#-deib-assessment) - [✅ Logical validity](#-logical-validity) - [🛡️ DARVO analysis](#️-darvo-analysis) - [🎯 Conclusion](#-conclusion) - [🔄 Refactor appendix](#-refactor-appendix--adobe-has-a-big-problem) ## 📝 Summary - **What happened.** Syntax, the podcast hosted by Scott Tolinski and Wes Bos, published a 35:08 episode (uploaded 2026-10-14, episode 1047) about a third-party project that claims to have clean-room ported seven Adobe apps to Rust. The hosts say the project appeared "yesterday" relative to recording. One host installed it, opened his own Photoshop files, and reports masks, blend modes, filters and text rendering matching Photoshop, with one known miss: drop-shadow texture (00:05:04). - **The claim.** Two. First, the port is good enough to take seriously: it opens faster than Photoshop and handled a stack of eight effects with blend modes and a mask. Second, copying Photoshop is not the real threat to Adobe. The host argues the next replacement will be "a new way of working", the way Figma displaced Photoshop for web design. - **What the hosts say is missing.** Multiple artboards (a roughly 2 GB document with about seven artboards crashed the app), collapsible effects in the side panel, Control-click layer masking, and drag-and-drop of an image onto an open document (00:12:09 to 00:13:36). The hosts frame UI interaction as the weak point of AI-generated software. - **Why it matters.** The episode ties three threads together: how fast a vibe-coded clone got close, Figma's allowlist approach to MCP clients, and what stays defensible when the code is cheap: interaction design and a maintainer who keeps caring after the first two days. - **Key stakeholders.** The unnamed port's maintainer and the company behind it (captions render the company as "Story Told or Storycraft"), Adobe, Figma and its spokesperson Dylan, Anthropic (named in a quoted clap-back), Blackmagic's DaVinci Resolve, Sentry (sponsor), and designers locked into Photoshop file formats. - **Risks and tradeoffs.** The port's legal status is not established in this source: the host says "legally dubious" for reverse engineering and says he doesn't know whether the project's agents were actually separated (00:07:55). The "70% to 80% of the Adobe suite" figure is a spoken estimate from commit history, not a measurement. The Sentry read at 00:24:56 is a paid placement. - **Bottom line.** The demo is evidence that a clone reaches file-format parity quickly. The hosts' own conclusion is that parity is not the endpoint. A product wins on UI craft and sustained maintenance, and Adobe's exposure is to a different tool, not to a cheaper copy. ## 📖 Main content **The opening claim.** The host presents a project that ports Photoshop, Illustrator, Premiere, Lightroom, InDesign and After Effects to Rust. The captions garble several names here ("Print. What is that? Inesign"); the hosts are listing Adobe's suite and the description says seven apps. He opens by pre-empting the "AI bro" reaction. He has used Photoshop for 25 years, has tried Affinity and other alternatives, and says none came close. He still pays about $80 a month for the suite, a figure he states without a source (00:00:43). **Why file compatibility is the real moat.** The hosts argue the industry standardizes on Photoshop's file format, so an alternative has to open those files with smart objects, filters and text intact. They say GIMP has existed for 28 years but fails on smart objects, many filters and text (00:09:19). Scott's side of the conversation adds that Figma is not a raster tool, so there is no web-design-grade substitute for blending modes, textures and print work. He says his own workflow changed to avoid raster editing after 2017 (00:20:56). **What clean room porting means.** The host's explanation: one agent studies the application and writes a description of every surface and feature, tests are written from that description, and a second agent implements from the tests without ever touching the first agent's source. The separation is what makes it a clean room. He adds that thousands of tests run against thousands of Photoshop files, and he doesn't know whether the project followed the separation strictly (00:07:13 to 00:07:55). This is the contested point of the episode. Clean-room engineering has a legal history that depends on documented separation, and no documentation of it appears in the transcript. **How the hosts handle early bugs.** One host warns against dismissing the project over one or two bugs: the code is two weeks old, so it should improve quickly (00:10:02). He also notes the project's builder obviously knows Rust and the apps. The other host adds the counterpoint that anyone can vibe-code a clone with 10% of the features. **Figma and MCP access.** The conversation turns to Figma's decision to approve MCP clients individually. The hosts report a case where a coding tool called Pi (captions: "PI") was blocked, the community contact pointed to a Google form, and the form returned "not accepting any at this time" (00:15:05). They quote Figma's Dylan questioning Anthropic's own closed ecosystem and agree the point is fair but doesn't address the block. Their broader read: platform owners treat agent access like the API era, where "we are not just a utility" meant routing users through the platform (00:16:32). **What replaces Photoshop.** Scott argues the replacement is not a free or faster Photoshop. He walks the history of web design tools: Photoshop, then Illustrator, then Sketch, then Figma, with Figma's multiplayer editing and browser delivery as the step change (00:18:45 to 00:20:56). Adobe adding AI to its own apps means "Photoshop with AI" is not a differentiator. **Moats.** Two named: interaction detail (click, drag, hold a key, right-click, modes, "state machines all the way down") and a maintainer who cares past the first two days. A bug filed in the project's Discord ("audio keeps crackling" for the video app) prompts a riff on low-level video bugs, which FFmpeg-style primitives only partly solve because hardware varies (00:23:05 to 00:24:32). That leads into the Sentry sponsor read, which argues that capturing errors with device context is the prerequisite for any agent to fix them. **Copying versus inventing.** The closing argument: copying is easy, inventing the next interface is the work. DaVinci Resolve is the positive example, praised for a steady cadence of useful releases (00:27:25). The last six minutes are sick picks (a 3D-printed server rack, an Insta360 camera) and are off-topic for the main argument. **Uncertain or unsourced.** - "70% to 80% of the suite" and the two-week build window come from commit history read on screen, not from a published project claim. - Neither host names the project or maintainer clearly. Captions give "Echelon Brandon Thomas on GitHub", which looks garbled. Check the episode links at syntax.fm/1047. - Whether the clean-room separation was real is stated as unknown by the host himself. - The DJI import-ban remark (00:33:14) and the Costco purchase limit (00:33:56) are unsourced asides. ## 🗒️ Notes #### ⏱️ [00:00:00](https://www.youtube.com/watch?v=SL8LD9oxVlY&t=0) Welcome to Syntax - Cold open starts mid-claim: a company "clean room ported" the Adobe suite. The 12-second intro chapter holds only the first sentence. #### ⏱️ [00:00:12](https://www.youtube.com/watch?v=SL8LD9oxVlY&t=12) Adobe is in trouble: a clean room port of the entire suite - Apps named: Photoshop, Illustrator, Premiere, Lightroom, InDesign (captions: "Inesign"), After Effects. The host says "vibe slop forked" and expects to dismiss it. - Credentials offered: 25 years on Photoshop, tried all alternatives, wants to stop paying $80 a month. The $80 figure is unsourced. - Tests on his own recent Photoshop files. Result: it opens faster than Photoshop and feature parity is incomplete. > [!quote] [00:01:29] > I'm shocked at how good and fast this is. - The framing question for the episode: should you vibe-code an exact copy, or is the next iteration of software something different? #### ⏱️ [00:02:22](https://www.youtube.com/watch?v=SL8LD9oxVlY&t=142) Scott's long-running Adobe grudge since 2017 - Scott says he published a video on July 6, 2017 on the LevelUp Tutorials channel titled how to mostly ditch Adobe, and hasn't opened Photoshop since 2017. - Triggers he lists: always-running Creative Cloud, updates, login, and dark patterns around cancellation. - His replacements: DaVinci Resolve for video, Figma over Illustrator. Photoshop never had a replacement he liked (Affinity Photo installed "forever"). #### ⏱️ [00:03:42](https://www.youtube.com/watch?v=SL8LD9oxVlY&t=222) Figma is not a raster tool: why every alternative falls short - Figma handles web design but has no raster blending equivalent. Texture work (the Syntax site, t-shirt designs) doesn't look right as vector. - Print: the Syntax skateboard box was designed in Illustrator. The host says Figma "spins away and crashes" on print design. - He asserts that anyone claiming better alternatives isn't a professional user. That's a sweeping claim with no evidence offered (see Logical validity). #### ⏱️ [00:04:40](https://www.youtube.com/watch?v=SL8LD9oxVlY&t=280) Opening real Photoshop files: masks, halftones and blend modes - Test file: a custom halftone built from eight stacked effects (patterns, gradient maps, levels, black-and-white variants, random variants), each with its own blend mode, one layer masked. It rendered correctly. - A Mad CSS thumbnail with several masks and drop shadows matched except for the drop-shadow texture. - The commit history on GitHub stops about two weeks before recording, which the host reads as roughly 70% to 80% of the suite in two weeks. The percentage is his estimate. #### ⏱️ [00:06:30](https://www.youtube.com/watch?v=SL8LD9oxVlY&t=390) What clean room porting actually means - Explanation: a spec agent describes surfaces and features, tests follow, and a separate implementation agent writes code it never saw the original for. > [!quote] [00:07:13] > what a clean room does is that they have one agent that is simply going through the entire application - Whether the project followed this is unverified by the host: "thousands of tests and thousands of Photoshop files" are cited as the test base. - Host aside: reverse engineering is "legally dubious", and a side reference to decompiling video games. #### ⏱️ [00:09:54](https://www.youtube.com/watch?v=SL8LD9oxVlY&t=594) A familiar interface, entirely built in Rust - The industry runs on Photoshop's file format, so a viable alternative must open those files. > [!quote] [00:09:19] > you have to use something that supports Photoshop files. - GIMP's 28 years are cited as insufficient: no smart objects, many filters missing, text broken. - The UI is also built in Rust, not on GPUI by the host's check, but on its own crate with tests for building the UI. #### ⏱️ [00:10:15](https://www.youtube.com/watch?v=SL8LD9oxVlY&t=615) Don't trash the whole thing over one or two bugs - Plea against dismissing a two-week-old project over early bugs. The pace implies fast improvement. > [!quote] [00:10:43] > don't throw the baby out with the bath water. - Maintainer is described as clearly skilled in Rust and in these apps. Captions garble the name and the company ("Story Told or Storycraft"). The host says the company builds GUI tools for LLMs in image editing and predicts they'll reimplement Adobe's generative features (circle an area, ask for a cowboy hat) on their own service. #### ⏱️ [00:12:14](https://www.youtube.com/watch?v=SL8LD9oxVlY&t=734) What's missing: artboards and the tiny UI interactions - Missing or broken: collapsible effects in the side panel, Control-click between layers to create a mask, drag-and-drop opening an image as a new file. - A roughly 2 GB document with about seven artboards and eight thumbnails crashed the app. The host guesses multiple artboards aren't supported. - Stated pattern: UI is the hardest part of vibe-coded software, despite hundreds of UI tests in the project. #### ⏱️ [00:13:53](https://www.youtube.com/watch?v=SL8LD9oxVlY&t=833) Where are the vibe-coded Photoshops? Four months later - Reference to a blog post titled "where are all the vibecoded Photoshops" from four months earlier asking if anyone ships professional software this way. The hosts treat this project as the answer. #### ⏱️ [00:14:33](https://www.youtube.com/watch?v=SL8LD9oxVlY&t=873) Figma's locked-down MCP and the backlash - Reported case: Pi blocked by Figma's MCP, support pointed to a Google form, form said not accepting submissions. - Scott has Figma history: he says he made early explainer videos for them and had a job offer around 2013 to 2015 (he's unsure which year). > [!quote] [00:15:47] > they basically have been taking a stance that's like kind of a lock down approach to AI tools. - Scott wonders whether Figma has a moat here, then concedes Figma is too complex to fork in a week. #### ⏱️ [00:17:22](https://www.youtube.com/watch?v=SL8LD9oxVlY&t=1042) Dylan's clap back at Anthropic's own closed ecosystem - The other host compares Figma to Amazon blocking shopping agents and platforms blocking AI readers, like the API era eight years earlier: platforms want users routed through them for experience and revenue. > [!quote] [00:17:16] > can you help me understand why you don't make various anthropic resources available via the MCP? - Hosts agree the jab lands on Anthropic (subscription limits on third-party harnesses are mentioned) but doesn't answer why Pi was blocked. #### ⏱️ [00:18:08](https://www.youtube.com/watch?v=SL8LD9oxVlY&t=1088) The real Photoshop replacement will be a new way of working - Photoshop's stickiness: nobody is happy to use Adobe, it's just the industry standard. - History cited: Photoshop, then Illustrator ("no fun"), then Sketch (Mac only), then Figma. > [!quote] [00:19:29] > I think whatever we end up landing on is not going to be the exact same thing of Photoshop but free - Adobe is adding AI to its own products, so "ours has AI" won't move people. The best working video-editing setup so far is MCP, per the host. - Scott's workaround for Photoshop: he stopped doing raster photographic editing. The other host answers that style follows tooling. #### ⏱️ [00:21:35](https://www.youtube.com/watch?v=SL8LD9oxVlY&t=1295) Good UI is still the moat - Moat one: interaction detail (click, drag, paste, modes). Figma won initially on this. > [!quote] [00:21:38] > with any of these AI generated apps, the weakest point by far is always the UI. - Complexity named: modifier keys plus hover plus right-click create modes. "State machines all the way down." #### ⏱️ [00:22:28](https://www.youtube.com/watch?v=SL8LD9oxVlY&t=1348) Caring about your project past two days - Moat two: someone who cares. Scott quotes the other host's own tweet from an hour earlier. > [!quote] [00:22:21] > The only moat left is caring about your project past two days. - The host filed the multi-artboard crash in the project's Discord and suggested adding the file to the project's test fixtures. #### ⏱️ [00:23:05](https://www.youtube.com/watch?v=SL8LD9oxVlY&t=1385) Audio crackles and paper cuts: the bugs only humans chase - A Discord user reports crackling audio in the project's Premiere alternative (captions: "FilmCraft"). > [!quote] [00:23:48] > Audio crackles, frames dropping, all of this type of stuff. - Low-level primitives like FFmpeg solve much video playback but hardware variance remains. Adobe has staff whose job is hunting hardware-specific bugs. The hosts' example is a rugged laptop with 11 layers and a mask. - "Seven or eight little paper cuts" and users quit. #### ⏱️ [00:24:56](https://www.youtube.com/watch?v=SL8LD9oxVlY&t=1496) Brought to you by Sentry (sponsor read) - Paid placement for Sentry, ending with the call to action at sentry.io (the description gives sentry.io/syntax). - Pitch: with agents fixing bugs, capturing errors, version and hardware context is the bottleneck. The host says he wished the crashed app had Sentry built in rather than hand-filing a report. #### ⏱️ [00:26:28](https://www.youtube.com/watch?v=SL8LD9oxVlY&t=1588) Copying is easy: the hard part is inventing what comes next - Closing argument against stopping at parity. > [!quote] [00:25:58] > copying this stuff is easy. - The host wants out of the $80 a month plan for something he uses four or five times a month, but says the winner will be someone constantly asking how to make it better. #### ⏱️ [00:27:28](https://www.youtube.com/watch?v=SL8LD9oxVlY&t=1648) DaVinci Resolve keeps shipping useful releases - Resolve is the positive example. The host has used Final Cut Pro, Final Cut Pro X and Premiere and says none ships useful releases at Resolve's rate. > [!quote] [00:26:42] > I don't want a Premier equivalent. - Banter on slang ("goated", "six seven") closes the section. #### ⏱️ [00:28:26](https://www.youtube.com/watch?v=SL8LD9oxVlY&t=1706) Sick picks and shameless plugs - Scott: a 3D-printed server rack (the KWS V2 design per captions), PETG filament, planned for Raspberry Pis, a Mac Studio, a Mac mini and a Synology under his desk. He ties his color and screw choices to "taste as a moat". - The other host: an Insta360 Luna gimbal camera for B-roll, chosen over DJI's Osmo Pocket 4, which he says is banned in the US (unsourced). Detachable screen and 4K cropping are his reasons. - Scott's wife's YouTube channel (captions: "Phases FM") on dyslexia and learning disabilities gets a plug. Costco's purchase limit of five is mentioned as a joke. - The episode ends at about 00:34:40 with a sign-off. ## 📚 References The transcript itself contains no links; the following come from the video description or are primary sources for named entities. - [Syntax episode 1047 and show links](https://syntax.fm/1047): description's link hub for the episode. - [Sentry for Syntax listeners](https://sentry.io/syntax): sponsor link from the description. - [Syntax on YouTube: Adobe has a big problem](https://www.youtube.com/watch?v=SL8LD9oxVlY): the source video, 35:08, uploaded 2026-10-14. - Adobe Photoshop, Illustrator, Premiere, Lightroom, InDesign and After Effects ([adobe.com](https://www.adobe.com)): the products being ported. - The Rust port project: the transcript never names it clearly (captions: "Story Told or Storycraft", "Echelon Brandon Thomas"). Find it through the syntax.fm/1047 links before citing. - [Figma](https://www.figma.com) and its MCP server docs: the lockdown discussion at 00:14:33. - [Model Context Protocol](https://modelcontextprotocol.io): the protocol under the Figma and Anthropic exchange. - [DaVinci Resolve](https://www.blackmagicdesign.com/products/davinciresolve): the release-cadence example at 00:27:28. - [FFmpeg](https://ffmpeg.org): the low-level video primitive named at 00:23:48. - [Rust](https://www.rust-lang.org): implementation language; the hosts also mention GPUI as a Rust UI framework they checked for. - [Sentry](https://sentry.io): error monitoring product in the sponsor read. - [Insta360](https://www.insta360.com): camera sick pick (captions: "Insta 360 Luna"). - [GIMP](https://www.gimp.org): named as the 28-year-old Photoshop alternative that fails on smart objects. ## 🗣️ Paraphrase The hosts open with surprise. A company ported Adobe's whole creative suite to Rust, and when one host dropped his own Photoshop files into it, stacked effects, blend modes and masks rendered the way Photoshop renders them. It launched faster than Photoshop too. He knows the gaps: multiple artboards crash it, some interaction shortcuts don't work, and the drop-shadow texture is off. He asks listeners not to throw out the project over those, since it's two weeks old. On the legal question, he explains clean-room engineering as a split between a spec-writing agent and an implementation agent. He admits he doesn't know whether the project kept that split, so the legal footing is open. The conversation then widens. Figma's decision to approve MCP clients one by one is read as a platform protecting its ecosystem, in the same way big platforms treated APIs years ago. The hosts concede Figma's spokesperson made a fair point about Anthropic's own restrictions, then say it doesn't answer the block that started the dispute. Their thesis comes last. A free or faster Photoshop won't unseat Photoshop. The winner will be a different way of working, as Figma was for web design. Until someone finds it, the defensible ground is interaction craft and a maintainer who keeps fixing odd bugs like crackling audio. The sponsor read for Sentry argues that capturing errors with device context is what lets agents fix such bugs. The episode closes on the point that copying is easy and inventing is hard, then moves to gadget picks. ## 📰 Newsletter **A clone of Photoshop in two weeks is the easy part.** Syntax's 35-minute episode on a Rust port of Adobe's apps has a demo worth watching and an argument worth borrowing. The demo: one host threw his real Photoshop files at it and got matching masks, blend modes and text. The argument: parity isn't the prize. Interaction detail and a maintainer who stays past day two are what keep users. For WordPress folks, the Figma MCP allowlist fight is the part to read closely. If your tooling assumes agents can reach a design tool, check whose permission you need first. Another item for the docs and release crowd: the hosts' point about capturing device and version info with each crash report applies to every plugin support thread. Links: [the episode page](https://syntax.fm/1047), [the video](https://www.youtube.com/watch?v=SL8LD9oxVlY), [Sentry](https://sentry.io/syntax) (sponsor of the episode). ## ✅ Action items #### DevRel - Developer advocacy: write a short comparison of "port the product" versus "invent the workflow" using this episode's Figma history (Photoshop to Sketch to Figma) as the example. - Education: run one workshop on reading AI-built clones critically: what to test (file import, edge-case files, interaction modes) before judging. - Developer experience: when documenting any MCP integration, state the approval path for third-party clients up front, since the Pi block at 00:15:05 shows the cost of an opaque process. - Developer marketing: do not repeat "70% to 80% of the Adobe suite in two weeks" without a primary source. It's a spoken estimate. - Community management: add a bug-intake template that asks for OS version, chip and app version, which the hosts say maintainers always have to request (00:25:58). - Internal/external/cross-project: ask the port project's maintainers for their fixture-contribution process, since the host suggested adding the crashing multi-artboard file. #### OSPO - Policy and governance: decide the organization's position on clean-room claims made by AI-built forks. Require documentation of the spec/implementation separation before adopting one. - License compliance: check licensing and any Adobe file-format terms before adopting a port that reads PSD files. - Security management: file formats parsed by a new Rust implementation are an untrusted-input surface. Fuzz-test the PSD parser before use on shared files. - Software inventory: if the port enters any workflow, record it with its version and maintainer count, since the project's own commit history is two weeks deep. - Developer support: publish guidance on when a young project's bugs are tolerable and when a crash on a 2 GB file is disqualifying. - Community engagement: consider funding upstream fixture contributions for projects the team relies on. - Strategic alignment: track a KPI for "time to first usable release" versus "months of sustained maintenance" when judging AI-built tools. #### Community - Maintainers: take the "caring past two days" point as a rubric. Publish a maintenance cadence so users can judge it. - Contributors: add a "send us the file that breaks it" path, as the host did in the project's Discord. - Sponsors: when a sponsor read appears in an episode you cite, label it as paid. - Ecosystem: run a cross-project check on which tools gate AI agent access, since Figma's allowlist could repeat elsewhere. #### WordPress - Core: no direct action from this source. - Core Performance: no direct action. The "opens faster than Photoshop" claim is one host's observation, with no benchmark. - Core AI: note the Figma MCP allowlist when planning any agent workflow that depends on design-tool access. - **Core Program Team:** no direct action from this source; nearest relevance is cross-team awareness of third-party MCP access policies. - Mobile: no direct action from this source. - Playground: no direct action from this source. - Community: no direct action from this source; nearest relevance is the "caring past two days" maintainer point for meetup and WordCamp tooling. - Training: draft a lesson on what makes a tool defensible once code is cheap: UI craft, maintenance, and file-format compatibility. - Marketing: no direct action from this source. - Support: copy the host's list of crash-report fields (OS, chip, app version) into support-form guidance. - Design: audit block-editor interaction states the way the hosts describe modifier-key modes, since "state machines all the way down" applies to editor UI. - Documentation: add the failure-case list (artboards, masking shortcuts) as a model for documenting known gaps in any compatibility claim. - Photos: raster workflows are the topic here. If the team accepts PSD sources, test whether an open implementation reads them before relying on it. - TV: no direct action from this source. - Polyglots: no direct action from this source. - Accessibility: check any new design tool for keyboard operation and screen reader support before recommending it. The episode doesn't mention either. - Meta: no direct action from this source. - CLI: no direct action from this source. - Hosting: no direct action from this source. - Tide: no direct action from this source. - Openverse: no direct action from this source; nearest relevance is file-format handling for image ingestion. - Themes: no direct action from this source. - Plugins: no direct action from this source. #### The WPOCC - Programs: no direct action from this source. - Projects: no direct action from this source. - Funding experiments: if The WP Open Community Collective funds maintenance, the episode's "caring past two days" is an argument for funding duration over launch. ### FAIR - Technical independence: the Figma allowlist example shows how a platform can gate access to its data. Document which parts of the stack depend on one vendor's approval. - Policies and covenants: no direct action from this source; nearest relevance is openness of agent access to metadata. - Productized governance artifacts: add an "agent access policy" row to scorecards for tools that restrict MCP clients. ## 🧪 Hypotheses 1. A fixture set of 50 real Photoshop files with layers, smart objects and text reveals more parity gaps than the project's own tests, with at least 10% of files failing on first open. 2. Multiple-artboard documents are the single largest class of crashes for the port, accounting for most failures among files over 1 GB. 3. A maintainer's median time to fix a reported crash predicts user retention better than feature count, across AI-built forks. 4. Figma's MCP allowlist reduces the number of distinct agent clients using Figma by more than 30% compared with an open policy, without reducing paid seats. 5. Photoshop users who migrate to a raster alternative do so only when the alternative opens their existing files without loss, regardless of price. 6. Projects that attach device and version context automatically to crash reports close issues at least twice as fast as projects that collect it by hand. ## 📊 Methodology - **H1 and H2.** Collect 50 PSDs from volunteers, spanning layer count, smart objects, text and artboards. Open each in the port and in Photoshop, record pass/fail per feature class, and tally failures by file size band. - **H3.** Pull issue-tracker data for 20 AI-built forks. Measure median time to first fix after a report, then compare against monthly active users or stars retention over 6 months. - **H4.** Collect public lists of MCP clients that connect to Figma versus design tools with open policies. Compare client counts and seat growth over 12 months. Treat it as observational, with the caveat that other factors differ. - **H5.** Survey 100 working designers on what blocks switching, with price, file compatibility and UI familiarity as ranked options. Include only people who use Photoshop weekly. - **H6.** Compare time-to-close for crash issues in two matched sets of projects, one with automatic capture such as Sentry and one without, controlling for team size. - **Controls.** Pre-register thresholds. Record whether the host's "I don't know" about clean-room separation is resolved by project documentation. ## ❓ Questions - Does the port document its clean-room process, and would that documentation hold up to a legal challenge? - What is the project's name, license and maintainer count, given that captions mangle all three? - Where does the "70% to 80%" estimate come from, and what does it measure: apps, features or lines? - Do Adobe's file-format terms restrict independent implementations of PSD reading? - What is the next interface for raster and video editing that the hosts say nobody has found yet? - Would Figma's allowlist survive public pressure, and what criteria would the approved list use? - Does AI-generated UI code have a systematic weakness around modifier-key state, or does it just need more fixtures? - How should a maintainer budget attention for hardware-specific bugs like crackling audio across thousands of device combinations? ## 🔮 Future research - **Clean-room evidence for agent-built ports.** No standard exists for proving two agents never shared context. A reproducible audit trail would turn "we did it clean room" from a claim into a checkable artifact. - **File-format conformance suites.** PSD, AI and PRPROJ lack open conformance tests. A community-maintained suite would let anyone measure parity. - **UI interaction testing for generated apps.** The hosts say UI is the weak point. A benchmark of modifier-key and mode behavior would quantify it. - **Maintenance durability.** Track which AI-built projects are still shipping fixes at 6 and 12 months, to test the "caring past two days" claim. - **Platform access policies for agents.** Compare allowlist, open and paid-tier approaches to MCP across design, commerce and social platforms. ## 💡 Implications - **Practice.** File-format compatibility is the first test of any alternative creative tool. Teams evaluating one should bring their own awkward files, as the host did. - **Policy.** Clean-room claims need documentation. An organization adopting an AI-built port should ask for the separation records, since the legal comfort depends on them. - **Security.** A newly written parser for complex binary formats is an attack surface. Treat files from unknown sources as untrusted until the project shows fuzzing and review. - **Education.** The Figma history (Photoshop, Illustrator, Sketch, Figma) is a compact case study in how tools get replaced by new workflows, not copies. - **Funding.** If the scarce asset is a maintainer who cares past day two, funding should reward sustained maintenance and fixture contributions, not launch events. - **Vendor literacy.** The episode contains a paid Sentry read between two product arguments. Label sponsorship when citing the episode's claims. ## 🧠 Learning styles assessment - **Reading level estimate.** Estimated, not computed by a tool: a casual spoken-word transcript with short sentences. A Flesch-Kincaid grade of about 6 to 7 and Flesch Reading Ease of about 75 fits conversational speech. Domain terms (clean room, MCP, state machines, raster) raise the effective level to an intermediate developer or designer audience. #### 👀 Visual learners - Present: the host shows his Photoshop files in the port on screen, including the halftone stack and the Mad CSS thumbnail. The transcript refers to them with "this" and "here". - Improve: add a side-by-side of Photoshop and the port rendering the same file, with the drop-shadow difference marked. #### 🔊 Aural learners - Present: full conversational narration by two hosts, with a chaptered structure. - Improve: the episode is already audio. A 2-minute summary at the top of the sick-picks-free portion would help listeners skip to the argument. #### 🗣️ Verbal learners - Present: the clean-room explanation is a plain-language definition (00:07:13), and the Figma-history analogy does explanatory work. - Improve: define MCP and "state machines" when first used. Neither is explained. #### 🧩 Kinesthetic learners - Present: the host's own test is a model exercise: open your own files in the tool. - Improve: give the project's install steps and a short checklist of test files to try. #### 🧠 Logical learners - Present: premise-to-conclusion arc: parity is easy, so parity can't be the moat, so UI and care are. - Improve: the "70% to 80%" estimate and the "$80 a month" figure need sources to be extended or tested. #### 🧑‍🤝‍🧑 Social learners - Present: two hosts debating and agreeing, plus the Discord anecdote of filing a bug. - Improve: add a discussion prompt: what would your replacement for Photoshop do differently? #### 🧘‍♀️ Solitary learners - Present: short chapters allow self-paced replay of the moat argument at 00:21:35. - Improve: a reflection question: what does your tool do that a clone could not? #### 🌿 Naturalistic learners - Present: little explicit. The 3D-printed rack and color choices in the sick picks touch on materials (PETG and heat warping). - Improve: tie the maintenance argument to lifecycle: tools that last because people tend them. ## 📊 Accessibility assessment #### Screen reader semantics - Present: YouTube supplies landmarks and player controls, and 20 chapter markers give a navigable outline. - Improve: on-screen demos (the halftone file, the Discord post) carry information the audio only points at with "this" and "here". WCAG 2.2 AA 1.1.1 (Non-text Content) and 1.2.5 (Audio Description, Prerecorded) aren't met for that visual content. #### Keyboard navigation - Present: the player is keyboard operable by platform default, which covers 2.1.1. - Improve: nothing in the video adds a barrier. 2.4.7 (Focus Visible) is a platform property. #### Contrast and readability - Present: not verifiable from a transcript. - Improve: show app UI at readable sizes during the demo so 1.4.3 and 1.4.11 hold for on-screen content. #### Text resizing and reflow - Present: the player scales and captions resize independently. - Improve: burned-in text in screenshots won't reflow, which touches 1.4.4 (Resize Text) and 1.4.10 (Reflow). #### Media and captions - Present: auto-captions and chapter markers exist, and they made this analysis possible. - Improve: captions garble the project and maintainer names, "Inesign", "Pi" as "PI", and "Photopea"-like words such as "photograph". Corrected captions would satisfy 1.2.2 (Captions, Prerecorded) in substance. The sponsor read isn't labeled in captions. #### Forms and interactions - Present: none in the video. The sponsor URL is spoken and listed in the description. - Improve: keep the link as selectable text in the description, which already works. 3.3.2 applies to the landing page, not the video. #### Standards reference - Gaps against WCAG 2.2 AA: **1.2.2** (caption accuracy on proper nouns), **1.2.5** (no audio description of on-screen demos), **1.1.1** (visual evidence not described), **1.4.4 / 1.4.10** (burned-in text). The caption fix is the cheapest and highest impact. ## 🌍 DEIB assessment #### Cultural breadth - Present: references span a Canadian host, US-centric tools and platforms, and an open-source Rust project. One host notes he lives in Canada when discussing a device ban. - Improve: the discussion centers on US and Canadian tools. No view from designers using Photoshop in other markets, where $80 a month weighs differently. #### Inclusive language - Present: casual, second-person, and no slurs. "AI bros" is used as a pejorative for a group of speakers. - Improve: replace "AI bros" with a description of the behavior (people who announce one tool killed another without using either). One sick-pick joke about "my rack" relies on innuendo that some listeners will find off-putting. #### Neurodiversity support - Present: 20 chapters with descriptive titles help skip and revisit. Each section has a single topic. - Improve: the episode wanders (slang tangent at 00:27:28, camera banter) and mixes jokes into explanation, which can make the main thread hard to track. A one-line recap before the sponsor would help. #### Representation - Present: two hosts, and a plug for a creator who discusses dyslexia and learning disabilities (Scott's wife, a doctor of educational psychology). - Improve: no voices from working Photoshop users outside the hosts, or from the port's maintainer. A guest would add another view. #### Psychological safety - Present: the hosts invite readers not to dismiss early bugs, and they admit uncertainty ("I don't totally know"). - Improve: the sweeping line that critics aren't "professional" users can make viewers who use alternatives feel unwelcome. State the claim as a personal experience. ## ✅ Logical validity #### Strong reasoning - **Causal, 00:21:38 to 00:22:21.** The argument that UI interaction is the weak point in AI-generated apps comes with a mechanism: modifier keys, hover and right-click combine into modes, which makes "state machines all the way down". That explains why tests alone don't close the gap. - **Analogical, 00:18:45 to 00:20:56.** Photoshop to Sketch to Figma is a relevant precedent for tool replacement by workflow change, not by clone. Figma's multiplayer and browser delivery are concrete features, not vague "better". - **Direct evidence, 00:05:04.** The host tested a real file with eight stacked effects and reports one specific miss (drop-shadow texture). A named failure alongside a named success is credible. - **Self-limiting disclosure, 00:07:55 and 00:12:52.** The host says he doesn't know whether the clean-room separation was real, and lists what breaks (artboards, masks). Those concessions raise the credibility of the rest. - **Claim-evidence match, 00:10:02.** "Two weeks old, so improvable" is a fair expectation, though the hosts don't show a bug-fix rate. #### Fallacies - **No true Scotsman, 00:04:21.** "anybody who tells you that there is better alternatives is not actually a professional user of these types of things." Any professional who disagrees is defined out of the group. *Fix:* say what workflows fail in the alternatives and name them. - **Hasty generalization, 00:05:04.** A handful of one user's files stand in for "most of the Adobe suite". *Fix:* state the sample and report failures per file class. - **Unsupported quantification, 00:05:47.** The "70%, maybe 80%" figure comes from commit dates, not a feature count. *Fix:* publish a feature matrix or drop the percentage. - **Hyperbole and internal tension, 00:22:21.** "The only moat left is caring about your project past two days" conflicts with the two moats named a minute earlier (UI craft and care) and with Scott's later point about taste. *Fix:* say "one of the few moats". - **Tu quoque (reported), 00:17:16.** The quoted clap-back at Anthropic answers a blocked-client complaint by pointing to Anthropic's own restrictions. The hosts note it doesn't address the block. *Fix:* the point is fair as criticism of Anthropic, but a separate claim. - **Undisclosed or unlabeled interest, 00:24:32.** The Sentry read follows a bug-fixing argument without a transition marking the ad, and Sentry is named as the natural fix for the problem just discussed. The sponsor is a commercial interest in the episode. The chapter title does flag "sponsor read", which mitigates it. *Fix:* a spoken "this is a sponsored message" before the pitch. - **Appeal to a personal credential, 00:00:43.** "I have been a Photoshop user for 25 years" supports credibility on Photoshop but not on legal status of clean-room porting. *Fix:* separate technical and legal claims, as the host mostly does. #### Tone of voice - Excited and conversational, with openings like "Adobe is in trouble" and "holy smokes". The register flips from enthusiasm to caution (at 00:10:02 and 00:12:09) when the hosts list what's broken. Scott's comments are more skeptical and historical, the other host's more hands-on. Humor runs throughout, especially in the ads and sick-picks, and sometimes cuts across the argument. The strongest content sits in the middle third, where the demo, bug list and moat argument appear. ## 🛡️ DARVO analysis Not applicable to this source. DARVO (Deny, Attack, Reverse Victim and Offender) describes responses to accusation. This is a tech podcast discussing a product and a platform dispute, with no accusation against the hosts. - **Deny:** nothing is denied. The host admits uncertainty on legality and lists bugs. - **Attack:** the hosts criticize Figma's MCP policy and mock "AI bros", but this is criticism of a policy and a pattern of behavior, not an attack on an accuser. The reported clap-back from Figma's Dylan at Anthropic is a reverse-the-question move in the Figma-versus-community dispute, relayed secondhand, and the hosts say it doesn't answer the block. - **Reverse victim/offender:** no such move appears. ## 🎯 Conclusion The hosts demonstrate that a Rust clone of Photoshop can open real files with stacked effects, blend modes and masks in two weeks, and the same project fails on multiple artboards and some interaction shortcuts. Their durable argument is that parity isn't the endpoint: UI craft and a maintainer who keeps going after the first two days decide who wins, and the next Photoshop will be a new workflow in the way Figma was. Open points: the clean-room claim is unverified in this source, the "70% to 80%" estimate is a spoken guess, the Sentry read is a paid placement, and the project's name and maintainer are garbled by the captions. **Next steps for readers:** open your own difficult files in any alternative tool before judging it. Check the project's licensing and clean-room documentation through the links at syntax.fm/1047 before adopting it. And when an agent platform blocks a client, ask for the approval criteria in writing. ## 🔄 Refactor Appendix — Adobe has a big problem > Density: 2–5 Literature atoms (H3 pad), 0–3 Permanent candidates (H4 pad). > Emitted: 5 Literature, 2 Permanent. Resonance filter applied. The sick-picks segment > (00:28:26 onward) was dropped as off-topic. > > Konik claim-header rules: present tense, declarative, ≤10 words, searchable, unique. > > Atomicity: one declarative sentence per atom. "and"/"also" → split. > > Tag pool: #bmf · #addendum · #xref · #articleseed · #storystem · #share · #aphorisms ### 📚 Literature Split Pad #### [H1] Sustained maintenance is the remaining software moat > [!quote] [00:22:21] > The only moat left is caring about your project past two days. **Context & use**: - 00:22:21, Scott reading the other host's tweet from an hour earlier. The hosts pair it with interaction detail as the other moat. - Hyperbolic ("only"), so record it as a claim to test, not a finding. **Tags**: #xref #aphorisms #articleseed <!-- atomicity: 1 idea ✓ --> #### [H2] Clean room means spec and implementation agents never meet > [!quote] [00:07:13] > what a clean room does is that they have one agent that is simply going through the entire application **Context & use**: - 00:07:13, the host's explanation. He says at 00:07:55 that he doesn't know if the project did this. - Use as a definition of the method, and keep the "unverified for this project" caveat attached. **Tags**: #xref <!-- atomicity: 1 idea ✓ --> #### [H3] A free copy of Photoshop will not replace Photoshop > [!quote] [00:19:29] > I think whatever we end up landing on is not going to be the exact same thing of Photoshop but free **Context & use**: - 00:19:29, after the Photoshop to Sketch to Figma history at 00:18:45. - The claim is a prediction. Its evidence is one precedent (Figma), so scope it as analogy. **Tags**: #xref #articleseed <!-- atomicity: 1 idea ✓ --> #### [H4] Interaction design is where AI-built apps fail first > [!quote] [00:21:38] > with any of these AI generated apps, the weakest point by far is always the UI. **Context & use**: - 00:21:38, with the mechanism at 00:22:21: modifier-key modes, "state machines all the way down". - Matches the bug list at 00:12:09 (collapse panels, Control-click masking, drag and drop). **Tags**: #xref #share <!-- atomicity: 1 idea ✓ --> #### [H5] Figma approves AI tools for MCP by hand > [!quote] [00:15:47] > they basically have been taking a stance that's like kind of a lock down approach to AI tools. **Context & use**: - 00:15:47, summarizing the Pi block and the Google form at 00:15:05. Reported secondhand from social posts, not documented in the video. - The hosts treat this as the same pattern as API gating years earlier. **Tags**: #xref #addendum <!-- atomicity: 1 idea ✓ --> ### 🧠 Atomic Split Pad #### [P1] Parity is easy, so the moat is craft and care **Claim** (one sentence, Courtney's voice): When a clone can reach file-format parity in two weeks, what stays defensible is interaction detail and the maintainer's long-term attention. **Evidence pointer**: [[#H1]] · [[#H4]] **Scope & conditions**: Holds for software with dense interaction states such as editors. Weaker for products whose value is data or network effects. The hosts' evidence is one demo and two named bugs, not a measured sample. **Links**: - [[Software moats after AI codegen]] — because extends - [[Vendor benchmarks without methods]] — because contrasts on evidence quality — this episode's estimates are spoken, not measured - [[Open source maintainer burnout]] — because applies-in sustained maintenance <!-- atomicity: 1 idea ✓ --> #### [P2] A clean-room claim needs a record of the separation **Claim** (one sentence, Courtney's voice): An AI-built port is clean room only if the spec agent and the implementation agent's contexts are documented as separate. **Evidence pointer**: [[#H2]] **Scope & conditions**: Applies when legal exposure matters. The host explicitly doesn't know if this project did it. Does not settle whether a clean-room process is legally sufficient in any jurisdiction. **Links**: - [[Clean-room engineering]] — because extends - [[Agent provenance and audit trails]] — because applies-in AI-built forks - [[Open source license compliance]] — because evidence-for the need for adoption checks <!-- atomicity: 1 idea ✓ --> ### ✏️ Fleeting Queue - [F1] Find the Rust Adobe-suite project's real name, company and license. Captions give "Story Told or Storycraft" and "Echelon Brandon Thomas": https://syntax.fm/1047 #fleeting #inbox - [F2] Check whether the project documents its clean-room separation, since the host says he doesn't know (00:07:55) #fleeting #inbox - [F3] Open question: do Figma's MCP approval criteria exist in writing, and what happened to the Pi request? #fleeting #inbox - [F4] Article seed: "Parity is a two-week problem" on what remains defensible once clones are cheap #fleeting #inbox #articleseed - [F5] Collect awkward PSDs (smart objects, text, artboards) as a personal test fixture set for any raster alternative #fleeting #inbox ### → Things 3 (max 2 actions) - [ ] Look up the Adobe-suite Rust port from the syntax.fm/1047 links and note its license and clean-room documentation : research, 15 min #things - [ ] Build a 10-file PSD fixture set from my own work to test any Photoshop alternative : design tooling #things <!-- deep-read:sections end --> 📄 **Full text:** [[Full Document Contents/Articles/Adobe Has a Big Problem.md|Adobe Has a Big Problem by Syntax]] ## Highlights - [00:07:13] but like you shouldn't be able to like take, as much as you hate Adobe or whatever these companies are, you shouldn't be able to just crack it open and reverse engineer all of the things. Um, so what a clean room does is that they have one agent that is simply going through the entire application and figuring out what all the surfaces are, what all the features are, how they work, detailing absolutely everything about it, and then passing that description on to maybe writing tests, things like that, and then passing those on to an actual implementation agent. meaning that the one that is writing the code has never actually touched the the agent that is has been like kind of [00:07:55] going through and describing all of it and in that way they it's it's like like a clean room right you've never actually touched those two whether they did that or not here is I I don't totally know if that's the case but there is thousands of tests and thousands of Photoshop files that are are being tested against and then the same for the Illustrator, the video editing, the Premiere one, right? etc. All of those things. What is missing? Um, I have >> Wait, hold on. Hold on. Sorry. Sorry. Yeah. Sorry to interrupt. Can we just notice here what Wes is on screen is very similar to what I'm wearing right now. This is [laughter] >> It's a different t-shirt, but it's so close. Folks, ([View Highlight](https://read.readwise.io/read/01m4x7ch54enzpr24grgk67nbw)) - Note: [H2] Clean room means spec and implementation agents never meet - Tags: atom/literature/h2 - [00:15:47] might have been 2015 now that I think about it. Either way, they basically have been taking a stance that's like kind of a lock down approach to AI tools. And you got to wonder if the ability to have like no moat here in some of this stuff. Now granted, Figma is a complex application. You're not just going to slot fork it in a week or whatever, but like this is probably going to light some fire under some people. Like gosh, the the people are not happy with Figma's stance on this MCP thing. it it blew up for them. Um, it really really got a lot of people like like what are you doing? And and and Dylan responded to me and you know it was really great but like we have a different idea about MCP and the overall [00:16:32] ecosystem where basically we feel like we can control the ecosystem and I don't think you can you know >> like I also was was thinking a lot about this Figma stuff because like Figma doesn't want you to use the MCP. I'm I hate I hate when I read my own tweets, but like this is my thoughts of like Figma doesn't want you using the MCP because they want to be in charge. Amazon doesn't want you shopping with agents. Reddit doesn't want Claude doesn't let Claude access the threads. You know, like Twitter doesn't let chat GBD read tweets. >> Um there are big players and like this happened with APIs eight years ago where the big players are saying we are not just a utility. We don't want to lay down and just let you do whatever you want. We want you to go through us. Um ([View Highlight](https://read.readwise.io/read/01m4x7cpxnrzt35ypvj8vx6heb)) - Note: [H5] Figma approves AI tools for MCP by hand - Tags: atom/literature/h3 - [00:18:45] somebody just makes a free version of Photoshop we just keep on doing that right the replacement for Photoshop is going to be something that is a little bit different we saw that with Figma Right. Like we used to design websites in Photoshop and when we moved we moved over to to Sketch and then we moved >> well we moved to Illustrator first and that was no fun. >> Figma >> and then we moved to Sketch >> which only works on Mac still. >> Yes. [laughter] And it was because Figma came out with like a better way of doing it. It was it was a little bit to like rewire your brain as to like how things work and and whatnot. But it was just a totally not totally new but a new enough [00:19:29] evolution of software that people moved to that. And I think that whatever the next version of Photoshop is going to be and believe me there's a lot of people trying to figure out what is the next version of video editing, photo editing, raster. Everybody is trying to crack this right now. Nobody has cracked it just yet. It seems to be like the best thing we have is just MCP with video editing so far. Yeah. >> But >> I think whatever we end up landing on is not going to be the exact same thing of Photoshop but free or but faster or but a little bit better. Uh or like a lot of people right now are making software but with AI. I don't think that that's that's the move. It's got to be some generational shift that is enough for ([View Highlight](https://read.readwise.io/read/01m4x7ckyxr6gfatf4tj13xqja)) - Note: [H3] A free copy of Photoshop will not replace Photoshop - Tags: atom/literature/h3 - [00:21:38] like what's your moat? What's the moat that that is there? I think like two things right now. One, the UI really good interaction where somebody cares about the itty bitty clicking, dragging, pasting, all of those types of things. That is that's why Figma won initially and and I think that whoever cracks really good UI for a photo editing program with AI is is going to be the winner. So with any of these AI generated apps, the weakest point by far is always the UI. not just what it looks like, but the interaction, which is like what happens when you click and drag and hold a key and then >> like hover over top of something and [00:22:21] then right click and you're in a certain mode. Like this stuff is so so complex. So many state machines all the way down. So whoever can crack that and and also my the second thing that I think is a mode is is someone that just cares. [laughter] >> I was going to say Wes, you say you don't want to talk about your tweets. I wanted to talk about your tweet from 1 hour ago. The only moat left is caring about your project past two days. [laughter] It is so true, man. Everybody can slop together something. And it is the person who is using it, getting down into the details, getting into how it feels, how it works, using it, like what it actually is, rather than just slopping something out and calling it a day. I ([View Highlight](https://read.readwise.io/read/01m4x7cerr9mjrdzfj56eeb5mz)) - Note: [H1] Sustained maintenance is the remaining software moat - Tags: atom/literature/h1