IssuesAllIssue 08 · 8 NovIssue 07 · 24 OctIssue 06 · 10 OctIssue 05 · 27 SepIssue 04 · 11 MayIssue 03 · 3 MayIssue 02 · 29 AprIssue 01 · 17 Apr
Rx AI Briefing

They Rebuilt Adobe. Now Question Every Application.

Issue 06 · Coverage 24 Sep–10 Oct 2026 · Issued 10 Oct 2026·15 min read

In ten days, one open-source project published early reimplementations of twelve professional applications: seven from Adobe, three from Microsoft Office, AutoCAD and Pro Tools. None is ready for real work. All are more functional than they had any right to be. When software becomes reproducible, code stops being the moat.

Audio edition · They Rebuilt Adobe. Now Question Every Application.
0:009:28
This week · three headlines and one listen

Read separately they are a viral repository, a design choice inside it and a run of market signals. Read together they describe one shift: software is becoming reproducible, and the scarce thing is now the trust and permission to run it.

Recommended listen
Welcome to Rome, on Engineering the Future — 6 minutes · open.spotify.com — a travel story about AI as a companion rather than another application. The experiential version of this issue’s argument. See the closing section.
The Brief · software is becoming reproducible

“An open-source, clean-room reimplementation of Adobe Photoshop in pure Rust.”

  • 30 Sep 2026 · first commit at 06:12 UTC · five Adobe-style repositories created that afternoon
  • 2 Oct 2026 · all seven Adobe-style applications released
  • 10 Oct 2026 · 39,000+ stars, 5,700+ forks and 1,150 commits for PhotoCraft alone
  • The maintainers’ own verdict: real Photoshop parity is “still well below 50%”
— the PhotoCraft repository and roadmap, github.com/storytold/photocraft. Counts read from the GitHub API on 10 October 2026. Stars measure attention, not adoption.

It is tempting to dismiss this as another viral “vibe-coding” stunt. That would miss the signal.

Software is not disappearing. It is becoming reproducible, agent-operable and increasingly temporary — and that changes what an organisation is paying for when it buys an application.

When software becomes reproducible, code stops being the moat. What remains is context, trust and the permission to act.

Open SourceSoftwareEngineering
Story 01 · ArtCraft

Glitchy, but real

A developer frustrated with Adobe rebuilt seven of its applications, then started on Office. Testers found the results immature. They also found them far more functional than they had any right to be.

A developer became frustrated with Adobe — and responded by rebuilding it. Brandon Thomas, the Atlanta-based founder of ArtCraft, says he “got bit by the dark pattern ‘cancellation fee’ one too many times.”

Not literally, and certainly not completely. But in under three days from first commit to release, ArtCraft published seven free, open-source applications modelled on Photoshop, Illustrator, Premiere Pro, Lightroom, Acrobat, After Effects and InDesign. Then it began doing the same to Microsoft Office, and then to AutoCAD and Pro Tools.

PhotoCraft, the Photoshop-style flagship, was created on 30 September. By 10 October it had accumulated more than 39,000 GitHub stars and 5,700 forks, along with 1,393 pull requests and 1,190 issues — an extraordinary response for software that its own maintainers describe as early alpha, and “not yet a Photoshop replacement for daily professional work.”

Gizmodo reports that Thomas built the suite with Claude Opus 5.5, in Rust; he told Hacker News, “We’re using vanilla Claude Code for ArtCraft apps.” The README calls the work clean-room: “implemented from public specs and observed behaviour only. No proprietary code, shaders or assets.” That is ArtCraft’s description of its own method, not an independent legal judgement.

Figure 1 · rebuilt from the GitHub API

Twelve reimplementations, four vendors, ten days

GitHub stars for every ArtCraft application published between 30 September and 10 October, with the product each is modelled on and the day its repository was created. Snapshot taken 10 Oct 2026, 15:32 UTC.

PhotoCraft · Adobe Photoshop · 30 Sep
39.9k
LightCraft · Adobe Lightroom · 30 Sep
8.6k
FilmCraft · Adobe Premiere Pro · 30 Sep
8.1k
PDFCraft · Adobe Acrobat · 30 Sep
7.4k
VectorCraft · Adobe Illustrator · 30 Sep
5.8k
EffectCraft · Adobe After Effects · 1 Oct
4.3k
WordCraft · Microsoft Word · 7 Oct
2.7k
DesignCraft · Adobe InDesign · 1 Oct
2.4k
CADCraft · Autodesk AutoCAD · 7 Oct
1.7k
GridCraft · Microsoft Excel · 7 Oct
1.6k
DeckCraft · Microsoft PowerPoint · 7 Oct
1.2k
SoundCraft · Avid Pro Tools · 7 Oct
1.2k
Stars measure attention, not adoption or product quality. They are the evidence that thousands of people looked at applications representing decades of accumulated development and could suddenly imagine them being reconstructed. Data: GitHub REST API, repositories of the storytold organisation.

Independent testers did not find fully formed Adobe replacements. They found something more provocative: software that was visibly immature, yet far more functional than it had any right to be.

Gizmodo’s Mike Pearl, who uses Photoshop “twice a year”, found that lasso, feather, paste, transform, levels and colour “were all where I expected them to be”. In his words, “it feels like using glitchy, but real, Photoshop.” Free Transform, though, was “an unpredictable mixed bag”.

The maintainers are blunter still. The README lists the biggest gaps as AI and generative features, about twenty missing tools, depth in typography and professional workflows, and plug-in compatibility. Every Photoshop menu item is wired to a command, “but that measures wiring, not behaviour.” The roadmap’s own assessment puts real parity “still well below 50%”, and scores whether “a professional could switch for daily work” at roughly 25–35%.

This is not evidence that Adobe has been replaced.

It is evidence that the distance between absurd idea and credible prototype has collapsed.

The story expanded almost immediately. ArtCraft also published WordCraft, GridCraft and DeckCraft: early attempts to recreate Word, Excel and PowerPoint. Paul Thurrott notes that they “aren’t as far along as the Adobe-inspired apps.”

The FUTC channel tested all three rather than relying on screenshots:

1

WordCraft

Looked convincingly familiar, but struggled with basic formatting and image layout.

2

GridCraft

Recalculated its charts automatically and, through its built-in MCP server, took instructions from Claude to add a sheet and a chart. Several controls were buggy.

3

DeckCraft

Created and edited slides through both the interface and Claude, but lagged throughout and crashed once on Present.

The reviewer’s advice for real work was mature LibreOffice, which “is going to get you a lot farther than downloading some vibecoded app that’s only a few days old.” He still called it “a pretty cool project”, and was impressed by how quickly the applications had become recognisable and agent-operable. A YouTube demonstration is a practical observation, not a technical audit — but it is a better test than a screenshot.

That is the right balance: do not confuse breadth with readiness — but do not confuse immaturity with irrelevance.

Adobe declined to comment to The Register; Microsoft acknowledged the inquiry and offered no further response. As of 10 October every repository is still live, and no takedown notice appears in GitHub’s public DMCA log.

The legal position is unsettled rather than hopeless. Andres Guadamuz, writing on TechnoLlama, concludes “very tentatively that ArtCraft may be on the right side of the law in the US”; the exposure he sees is in licence terms, patents and trademarks. Ars Technica adds trade dress — an interface that mimics Adobe’s too closely. And the lawyer Neil Brown told The Register it could still be “litigated, when the party most likely to feel aggrieved has deep pockets.”

If software functionality becomes easier to reproduce, every organisation should ask what it is actually paying for.

Previous assumptionGenerative question
Complex software is inherently scarceWhich parts are now inexpensive to regenerate?
The application is the productIs the application merely an interface over deeper capabilities?
Feature breadth creates defensibilityWhich features cannot be recreated from observable behaviour?
Users operate softwareWill agents become its primary operators?
Interfaces remain stable for yearsCould interfaces be generated for each role, task or moment?
Buying software transfers capabilityIs the organisation really buying trust, support and accountability?
Application portfolios are permanent assetsHow much of the portfolio records what was historically too expensive to rebuild?

This is not simply another build-versus-buy decision. It introduces a third option.

Until now
Buy, or build.
From here
Buy, build — or generate.
Questions to carry forward
  • If an application’s features can be reproduced, what is the customer really paying for?
  • What percentage of today’s application estate exists because replacement used to be too expensive?
AgentsArchitectureSoftware
Story 02 · PhotoCraft

The application becomes a control surface

The most consequential PhotoCraft feature may not be image editing. It is that every operation a person can click is also a command an agent can run.

PhotoCraft’s README describes it plainly: “Every menu item, tool and dialog runs a command from one registry of 500+ commands.” The interface, the command line, a JSON control channel and an MCP server all call the same commands, “so anything you can click, a script or an AI agent can do too.”

That makes PhotoCraft two products at once.

For a person
A familiar visual workspace.
For an agent
A structured capability environment, with every operation addressable.

Traditional applications were designed around humans navigating menus, panels and forms. AI was subsequently added through a chat box or an assistant button.

ArtCraft points towards the inverse architecture:

Structure
The underlying document, timeline or model remains structured.
Address
Every meaningful operation becomes machine-addressable.
Operate
The agent becomes the primary operator.
Inspect
The interface appears when human judgement or intervention is required.

The application does not disappear. It becomes an inspection and control surface.

Generative software
Permanent capabilities underneath. Temporary interfaces above. Intelligence moving between them.
Questions to carry forward
  • Should the next enterprise platform expose interfaces to people, capabilities to agents, or both?
  • If interfaces become temporary, what must remain permanent?
EconomicsSecurityEnterprise
Story 03 · The market

The cost of assurance rises

ArtCraft did not emerge in isolation. Capital is crowding into the companies that produce software, and a second market is forming around controlling what that software is allowed to do.

Several market signals point in the same direction as ArtCraft. CB Insights drew them together in two research notes, on 24 September and 1 October.

  • 3 Sep 2026Cyera completes its $1 billion acquisition of Oasis Security, an agent-identity company.
  • 8 Sep 2026Cognition raises at a $48 billion valuation.
  • 15 Sep 2026Factory raises at $5 billion — in CB Insights’ words, Cognition is valued at “nearly 10x” Factory in the same month.
  • 17 Sep 2026JPMorgan begins restricting some engineers’ Claude access to credentials and internal systems, Business Insider reports.
  • 22 Sep 2026Okta announces expanded agent-governance controls, including a kill switch that revokes active agent tokens.
  • 1 Oct 2026CB Insights: 10 of 12 leading private coding-agent startups share an institutional investor with at least four of the others.

Whatever the eventual winners, capital is behaving as though software production itself is becoming a strategic platform. At the same time, a parallel market is forming around agent identity, access and control.

Those two signals belong together. AI makes software and autonomous capability cheaper to create. Organisations then spend more to control what those capabilities can access and do.

Production
The cost falls.
Assurance
The cost rises.

Most enterprise applications combine four layers. Generative software may pull them apart.

If this direction holds, enterprise software may separate into four layers, each with a different source of enduring value:

1

Systems of record

Authoritative data, transactions and institutional truth.

2

Capability services

Reusable operations exposed to humans and agents.

3

Agent orchestration

Context, planning, delegation, memory and coordination.

4

Generated interfaces

Situational experiences assembled around a person and an objective.

Questions to carry forward
  • Which enterprise applications are durable systems — and which are merely historical interfaces?
  • When software can be generated, who tests it, owns it and accepts the liability?
  • Does the winning software company own the application — or the trusted context and authority beneath it?
What becomes scarce

Code becomes abundant. Eight things stay scarce.

If code becomes more abundant, value migrates towards what cannot be generated easily.

ScarceWhy it resists generation
ContextProprietary organisational knowledge and operating history
TrustConfidence that the system will behave reliably
PermissionAuthorised access to systems, data and money
DistributionThe established route to users and workflows
CompatibilityFaithful operation across complex existing environments
AccountabilitySomeone responsible when the outcome is wrong
TasteKnowing what should be created, not merely how
JudgementResolving ambiguity, consequences and competing objectives

The future software company may own less of the interface and more of the trusted capability beneath it. The future enterprise may buy fewer interfaces while governing far more capabilities.

What this means for engineers

From producing code to designing constraints, systems and evidence.

This does not mean engineering is over.

ArtCraft was not produced by an unskilled person typing a single prompt. Thomas describes himself as a fifteen-year industry veteran who has written “six-nines payment rails that move billions of dollars of transaction volume a day”, and the repositories reveal architectural choices, testing, file-format work, performance engineering and security considerations. He called the first release a “one-shot”; PhotoCraft’s history since then runs to 1,150 commits.

The interesting part is what his job has become. He has not “written a single line of code since February”, he wrote on Hacker News — what he needs now is to “quickly scan diffs and quickly see architecture at a glance.”

AI compresses implementation effort. It does not eliminate the need to decide.

Decisions AI does not remove
  • What architecture should survive constant generation?
  • Which behaviours require deterministic testing?
  • Where must compatibility be exact?
  • What should an agent be permitted to change?
  • How can generated code be inspected and maintained?
  • Who accepts responsibility for the finished system?

The scarce engineer may no longer be the person who can write the most software. It may be the person who can determine which software should exist — and prove that it can be trusted.

Executive response

Do not cancel the contracts. Interrogate the portfolio.

Executives should not cancel critical software contracts because an open-source alternative went viral. They should begin a more disciplined interrogation of the application portfolio:

  • Separate interface from value. Identify which applications are valuable because of their interface, and which hold durable data, controls, workflow logic or regulatory evidence.
  • Revisit vendor moats. Ask whether a supplier’s differentiation depends on unique capability, accumulated trust and network effects — or mainly on functionality that was historically expensive to reproduce.
  • Add “generate” to sourcing. Introduce generated and agent-operated software as a controlled option alongside conventional buying and building, initially in low-risk internal workflows.
  • Demand agent-ready architecture. Require structured commands, APIs, event histories and policy controls — not merely a chatbot embedded in the existing interface.
  • Govern proliferation. Assume cheaper creation will increase rather than reduce the number of applications. Establish ownership, testing, identity, permissions, lifecycle and retirement controls before generated tools spread.

Your application portfolio may be a historical record of what used to be too expensive to rebuild.

A point of view

Software is not over. Software as a fixed, scarce and permanent artefact may be.

On 1 October, Thomas wrote on Hacker News: “Software is over. But for real. I just one-shotted Photoshop.” Ars Technica put the first three words in its headline. He later said it was meant as “a warning to my fellow software engineers”, not a boast.

ArtCraft’s applications may stall. Some may be removed. Legal challenges may emerge. Professional users may reject them. Their present quality is not the point.

The point is that thousands of people looked at applications representing decades of accumulated development, and could suddenly imagine them being reconstructed. They did not rebuild Adobe perfectly. They rebuilt enough of it to make reconstruction plausible.

That changes the strategic question.

From
Can this application be rebuilt?
To
What remains valuable when rebuilding it becomes plausible?

The generative future is not a world without applications. It is a world in which applications become contestable, interfaces become situational, agents become operators — and human attention moves towards intent, judgement and accountability.

Code becomes abundant. Context, trust and permission remain scarce.

Recommended listen · and a complementary read

Engineering the Future · 6 minutes. A travel story about AI becoming a companion rather than another application. It is the experiential version of this issue’s argument: intelligence understands the journey, coordinates services and generates the right interaction around the traveller’s immediate context. Listen on Spotify.

Engineering the Future is the complementary read: Jack Wang’s August 2025 deck on how AI is changing engineering practice, from spec-driven engineering to brownfield estates, with the Rome episode at its centre. Read the deck.

Issue compiled 10 October 2026, covering 24 September–10 October. Figure 1 was rebuilt by Rx from the GitHub REST API (repositories of the storytold organisation, snapshot 10 Oct 2026, 15:32 UTC); commit, pull-request and issue counts come from the same API. Primary sources: the PhotoCraft README, roadmap and agent instructions; the ArtCraft applications page; Brandon Thomas’s posts on Hacker News between 28 September and 10 October, including the thread “ArtCraft Apps – open-source Adobe compatible suite written in Rust”; the Model Context Protocol.

News and analysis: Gizmodo, “Someone Vibe Coded a Free Knockoff of Adobe Creative Suite” (Mike Pearl, 7 October); PetaPixel, “Someone Rebuilt Free, Open-Source Versions of Photoshop, Premiere, and Lightroom Using AI” (Jeremy Gray, 7 October); Ars Technica, “‘Software is over’: Bold AI developer takes aim at Adobe with open source clones” (Kyle Orland, 7 October); Design Week, “First AI came and scooped up your art. Now it’s scooped up Adobe.” (David Coveney, 7 October); The Register, “Rust upstart reveals ‘reimplementation’ of Microsoft Office – with Claude on coding duty” (Richard Speed, 8 October); Computing, “This time, software really might be over” (Penny Horwood, 8 October); Tedium, “Let’s Deconstruct Photoshop” (Ernie Smith, 8 October); Thurrott, “Open Source AI Dev is Going After Microsoft Office, Too” (Paul Thurrott, 9 October); TechnoLlama, “Recreating Adobe: AI and the limits of software copyright” (Andres Guadamuz, 9 October).

Market signals: CB Insights, “Who’s on both sides? The investors backing rival coding agents” (Alex Lee, 1 October), “These startups are building the security layer for AI agents” (Adya Pandey, 24 September) and “Anthropic’s IPO filing maps the AI safety startup market” (Adya Pandey, 1 October); Cognition, “Do it all with Devin: Announcing our Series E” (8 September); Factory, “Factory raises $200M at $5B valuation” (15 September); Business Insider, “JPMorgan rolls out Claude changes: $2,000 spending limits and extra security” (Alice Tecotzky, 17 September); Okta, “New Okta for AI Agents innovations increase visibility into agent behavior, secure connections at runtime, and enforce continuous agent governance” (22 September); Cyera, “Cyera Completes Acquisition of Oasis Security” (3 September).

Practical testing: FUTC, “Somebody Vibe Coded Microsoft Office, and I tried it (so you don’t have to)” (9 October) and “Somebody Vibe Coded EVERY SINGLE ADOBE App” (7 October, testing three of the seven); Level Image, “I Tried The Free Open-Source Adobe Clones (Are They Usable?)” (8 October); RoboNuggets, “Someone Just Rebuilt Adobe for Free, and Claude Can Run Them All” (8 October); piauski, “The Adobe Slop Experience” (8 October).