If everyone can build, what is the value of software?
Just a couple of years ago, recreating a software ecosystem that took decades and millions of dollars to build would have been unthinkable. AI-assisted coding has now fundamentally changed that equation. The implications go far beyond software: they strike at the business model of every technology provider, including those in localisation.
The collapse of software’s technical moat
When LLMs first arrived as glorified chatbots. Their contextual understanding was fascinating, but they were constrained by their training limits. Software remained well protected, defended by moats of technical excellence that few could breach. Then, heavily funded, AI-enabled startups began producing new-generation products to solve complex problems.
Only a few people watching the market knew this was a short-term game. As models got better at understanding context, the quality of their code improved dramatically. Early "vibe coding" took significant time to turn into a finished product, and still relied on specialist front-end and back-end developers working around rudimentary tools. Today, those technical barriers have largely fallen away.
A small, agile team can now take a software problem and solve it astonishingly fast by reimplementing functionality rather than copying proprietary source code. Yet a working prototype is not a production-grade service. AI can generate functional interfaces, but maintaining secure, reliable, and legally sound software still demands rigorous engineering discipline.
This transition will take time while security practices and code verification mature. But as model providers build stronger vulnerability checking into their tools, small development teams will routinely replicate the software they need.
We saw a glimpse of this acceleration when Amazon rebuilt part of its Bedrock inference platform. Using its own agentic coding tool, Kiro, a team of six engineers finished in 76 days a project originally scoped for 30–40 engineers over 18 months.
ArtCraft and the rebellion against subscriptions
So, when ArtCraft released a free, open-source alternative to seven of Adobe's flagship applications, I was completely caught off guard; the project seemed to come out of nowhere. As the reactions rolled in, the underlying strategy became clear: this was not just a cool punk rock technical flex, but a direct challenge to the subscription model. By positioning itself as a champion of creators, fighting the cost of creation, ArtCraft instantly became a hero to anyone frustrated by the relentless monetisation of the creative industry.
Outside the big platforms, smaller teams are using similar tools to study how applications behave and build open alternatives in languages like Rust. Their goal goes beyond bypassing frustrating paywalls; it is a cultural statement about the freedom to build alternatives.
Consider how our relationship with creative tools has evolved. We moved from owning software outright to licensing it through subscriptions, paying monthly and finding essential features locked behind tiered pricing. Cloning is a direct response to that model. These open alternatives are free and maturing quickly, though they are not yet finished products: ArtCraft is still alpha software, and early reviewers found serious gaps. But once a project is in the hands of the open-source community, rapid iteration suggests that almost anything can eventually be replicated.
The rise of the software cloning movement
There was another open-source innovation that landed around the same time, an AI agent skill called REA, which stands for Reverse Engineer Anything. This essentially lets AI agents take apart software, down to its machine code, and rebuild how it works. In conjunction with the ArtCraft suit, a whole community of open-source developers and vibe coders start the “Craft” movement. Creating open-source clones of software and reverse engineering old and new hardware.
This has an impact on the world of enterprise where leadership is focused on efficiency and the bottom line. Operations and IT can now work together (which is harder than it sounds), deploying small AI-assisted development pods to build internal tools that solve internal problems and also looking at reducing SaaS overheads. Tools such as Claude and AI agent skills like REA let them look at services and software solutions and make a decision. If they know the input and process, they could build internal solutions for themselves, potentially removing vendors and SaaS from the equation.
Complex platforms with deep ecosystem lock-in should remain stable in the medium term. SaaS products that are essentially API/technology aggregators wrapped around workflows are highly vulnerable. Enterprises are no longer just evaluating third-party vendors; they are assessing their own capacity to build to replace.
The localisation sector is not immune. Pipelines already lean on AI for first-pass translation, but the drive now extends to replacing management software entirely, from asset tracking and workflow monitoring to AI-driven dubbing. If studios can run that tooling themselves, they will likely keep direct, tiered rosters of trusted experts on their own platforms, cutting out middlemen whose only value is connecting contractors to clients. LSPs that want to stay in the picture will need to offer something harder to replicate than matchmaking.
The TMS business model faces a reckoning
For years, translation management systems (TMS) have been the backbone of localisation operations. But the traditional model carries heavy trade-offs: expensive integrations, gatekept features, surcharges for AI, and dependence on vendor roadmaps.
That is why OpenLoc, announced by ModelFront in October, is so timely. Its code is released under the Business Source License, a source-available licence that converts to open source, and teams can self-host it for free. Positioned as an AI-first TMS, it offers flexible model choice and customisable workflows, and system-wide Model Context Protocol (MCP) support lets LLMs drive workflows directly. The business logic is telling too: ModelFront makes its money from AI that checks and fixes translations, so giving away the TMS turns the management layer into a distribution channel for its core service.
Consider traditional TMS economics: per-seat licensing, per-word processing fees and CMS integration costs. OpenLoc lets internal DevOps teams start from open code and build bespoke solutions without paying a middleman at every step. Nothing is stopping clients from running OpenLoc with a frontier model or internal models doing the linguistic checks instead; only time will tell.
Historically, localisation teams had two choices: buy off-the-shelf software that almost fit, or spend heavily building a custom TMS in-house. Today there is a third option: assemble. Localisation teams working alongside developers can use AI coding assistants to tie together open software, custom AI agents, and direct connections to linguists.
When software becomes a commodity, what has value?
As basic software interfaces become commodities, where does the real value lie? The defensible assets are the ones that are genuinely hard to replicate: trusted enterprise infrastructure, zero-downtime operations, hardened security, rigorous data privacy, accumulated translation memories and strict compliance. Human expertise matters the most, because AI alone cannot reliably replicate a linguist's deep understanding of cultural nuance. It may get there one day, but for now, it has no lived experience in a culture.
Ultimately, the question for vendors, LSPs, and enterprise teams is not whether AI can build a TMS or a software solution. It is this: if a feature becomes easy to reproduce and workflows can be assembled in days, which parts of your technology stack are you still willing to buy, and what will you now consider building and servicing it?
