For about five years, one acronym carried the weight of an entire architectural argument. If you were rebuilding an enterprise commerce stack and you were serious, you were going MACH. Microservices, API-first, cloud-native, headless. The alternative, a single monolithic suite, was treated as the thing you escaped, the legacy you apologised for. Then in March 2025 a publicly traded platform vendor that had paid to put its name behind the movement stood up and called it a path paved with hidden costs, operational nightmares, and financial ruin.
That was VTEX, and its exit from the MACH Alliance did not kill composable commerce. But it broke the spell. The interesting question is not whether MACH was right or wrong. It is why a genuinely good idea got oversold, what surfaced when real companies tried to live inside it, and where the consensus has landed now that the religion has worn off.
Where MACH came from
MACH was never just an engineering pattern. It was a coalition with a marketing budget.
The MACH Alliance was founded in June 2020 by commercetools, Contentstack, EPAM Systems, and Valtech, with ten more inaugural members at launch, among them Algolia, Contentful, and Fluent Commerce. The pitch was straightforward and, on its own terms, correct. Monolithic commerce suites like the older Magento, Salesforce Commerce Cloud, and SAP Commerce Cloud bundled everything into one codebase: cart, catalogue, search, content, promotions, checkout. That made upgrades painful, customisation expensive, and vendor switching close to impossible. You bought the whole thing and you lived with the whole thing.
The MACH answer was to break commerce into independently deployable services connected by APIs, each one a best-of-breed product you could swap without touching the rest. Want a better search engine? Plug in Algolia or Constructor. Better content management? Pick a headless CMS. The commerce engine, the storefront, the order management system, all separate, all replaceable. commercetools had built its business on exactly this model, with hundreds of granular APIs rather than one fused platform.
The argument was real, and for a certain kind of large, complex retailer it paid off. The problem was the framing. MACH stopped being one valid approach among several and became a purity test. Vendors advertised MACH certification. Agencies built practices around it. Gartner began noting that a growing majority of organisations would be steered toward composable digital experience technology rather than monolithic suites. If your stack was not fully decoupled, the implication was that you simply had not caught up yet.
That is the setup for a backlash. Not a bad idea, but a good idea sold as the only idea.
What broke in practice
The gap between the MACH pitch and the MACH bill showed up in a few specific, repeatable places.
The first is integration. A monolith ships its parts pre-connected. A composable stack does not. Every API boundary between the commerce engine, the search service, the CMS, the OMS, and the payment layer is a connection that someone on your team has to build, test, monitor, and repair. Practitioners describe drowning in API mediation layers, writing glue code whose only job is to make services that were never designed together behave as if they were. When that glue ages, you get the worst outcome in the category: a distributed monolith. You have paid for the operational overhead of many separate systems while still inheriting tight coupling, shared dependencies, and cascading failures. The common engineering view is that a composable architecture only avoids that fate when a dedicated platform team treats the integration layer as a product in its own right, with an owner, a roadmap, and a budget. Skip that, and the stack drifts back toward coupling on its own.
The second is cost, and specifically cost that does not appear in the sales deck. Best-of-breed means a separate contract, a separate licence, a separate support relationship, and separate infrastructure for each component. None of them looks expensive in isolation. Added up across development, integration, maintenance, and operations, they compound. VTEX's own argument is blunt on this: organisations effectively become software companies, and what looked cost-effective in evaluation becomes a financial black hole. The honest framing is that total cost of ownership, not the licence line, is the number that decides a commerce platform, and a composable build only wins that comparison when the integration and maintenance load is counted in from the start rather than discovered later.
The third is operational friction, which is the cost nobody models. When your commerce capabilities live in separate products, your business users live in separate interfaces. A merchandiser updates the catalogue in one tool, promotions in another, content in a third, search rules in a fourth. Each system has its own login, its own logic, its own training curve. The agility that composability promised on the engineering side can quietly disappear on the business side, where the people who actually run the store now spend their day switching tabs.
VTEX founder and co-CEO Mariano Gomide de Faria has used a single concrete example to put numbers on the trade-off. A retailer wanted to swap VTEX search for an external search product. Search, by his account, represents roughly 3 percent of total platform cost, yet running that one component as a genuinely independent service was going to take around 12 people to own the integration and operations. The retailer's response was to ask for a 30 percent platform discount to absorb the new burden. That is the composable bargain stated honestly: real flexibility, bought with real, recurring headcount.
What VTEX's exit actually signalled
On 26 March 2025, VTEX announced it was suspending its support of the MACH Alliance. This mattered for reasons beyond one company's membership status.
VTEX is not a fringe critic. It is a NYSE-listed commerce platform used by more than 2,400 brands, strong in Latin America and in enterprise marketplaces, and Gartner named it a Challenger in the 2025 Magic Quadrant for Digital Commerce. It had been a paying member of the Alliance. So when it walked, it was not an outsider sniping. It was an insider saying the emperor was underdressed.
Gomide de Faria framed the departure as a reaction against dogma. His position was that the pure best-of-breed MACH approach had created as many problems as it solved, that the movement had drifted from engineering pragmatism toward ideology, and that the industry should drop what he called composable extremism in favour of solutions that actually work for the business. VTEX laid out conditions under which it would return: lean and simple architecture, software ecosystems that connect out of the box, minimal or no middleware, and a posture that defends the interests of retailers and brands rather than the interests of technology vendors.
Strip away the rhetoric and the signal was specific. VTEX was not arguing that decoupling is bad. It was arguing that decoupling for its own sake is bad, that the number of independent services in a stack is not a measure of its quality, and that a movement run by vendors had developed a structural incentive to keep selling more components. The accompanying essay, which VTEX published under the idea of a MACH mirage, made the same case at length: composability is a tool, not a destination, and treating it as a destination is how companies end up with operational nightmares and a budget they cannot explain.
Importantly, VTEX did not retreat to the monolith. Its own positioning after the exit settled on a hybrid: pre-built, pre-integrated modules for things like cart, checkout, and payments, where speed and reliability matter more than customisation, and composable flexibility reserved for the areas where a brand is genuinely trying to differentiate. That is not anti-composable. It is composability with a cost discipline attached.
How the MACH Alliance responded
A weaker organisation would have either dug in or collapsed. The MACH Alliance did neither. It moved.
The immediate response from then-president Casper Rasmussen was to reframe rather than retreat. His line was that MACH represents guiding principles, not a rigid doctrine, that the Alliance welcomes healthy debate, and that the failures being described were failures of implementation rather than failures of the architecture itself. He pressed the idea of mutual accountability between business and technology leaders, and the point that architecture only counts if it is tied to real outcomes. Tellingly, both Rasmussen and Gomide de Faria agreed on one thing: MACH principles had become so widely adopted that they were no longer a differentiator. They were just how modern software gets built.
The deeper shift was already underway in the messaging. The Alliance had refreshed its brand in October 2024, moving from the bare four-letter acronym toward a broader set of MACH Principles, the explicit point being to clarify what MACH is and, just as importantly, what it is not. The talking point that emerged across 2025 was that the number of systems in a stack is not the measure of a good architecture, and that the goal is technology that is fit for purpose. That is a long way from the earlier mood, where more decoupling was simply assumed to be better.
Then the leadership changed. Rasmussen stepped down as president in September 2025, and the Alliance appointed Jason Cottrell, refreshing its executive board and, notably, adding end-user ambassadors rather than only vendors: among them the e-commerce technology lead at easyJet and the co-founder of a consumer startup. By the end of 2025 the Alliance reported more than 100 member organisations, its largest cohort, and had pivoted its public story hard toward agentic AI, positioning composable architecture as the necessary foundation for multi-agent commerce.
It also went looking for evidence. In February 2026 the Alliance published research, based on a survey of 600 enterprise technology decision-makers across seven markets, arguing that composability and AI returns are linked. The headline claim: 78 percent of organisations with fully scaled MACH technology reported clear ROI on their AI investments, against 13 percent of organisations still in early MACH planning, which the Alliance framed as a roughly sixfold gap. Treat a trade body's own research with appropriate caution. But the strategic intent is clear. The pitch is no longer be composable because monoliths are bad. It is be composable because that is what makes AI pay off.
Where the consensus landed
The backlash did not produce a winner. It produced a more honest question.
The mature position now, and you can hear it from Gartner, from practitioners, and increasingly from the MACH vendors themselves, is that composability is a spectrum, not a binary. The relevant question is not are you MACH or are you a monolith. It is which capabilities in your stack genuinely benefit from being independent, and which are fine bundled. Search, where best-of-breed quality is a real competitive edge, may be worth decoupling. Cart and checkout, where the requirement is mostly that they work reliably, often are not. The answer is per-component, and it depends on where a specific business actually competes.
The packaging is changing to match. Gartner's framing of packaged business capabilities, discrete and independently deployable units of functionality that a business user can recognise and consume whole, has become the practical middle path. Instead of buying a box of loose parts and an integration project, a company buys capabilities that arrive already connected. Vendors across the spectrum, VTEX with its pre-integrated modules being one example, now sell some version of this: most of the flexibility of composable, with much less of the glue work. It is composability as an outcome rather than composability as a construction kit.
For anyone planning a commerce build, the takeaway from the MACH backlash is not that the architecture was a mistake. Decoupling delivered real value for genuinely complex retailers, and it still does. The takeaway is to treat a vendor coalition's preferred architecture as an input, not an instruction. Count the integration headcount before you commit, not after. Decouple where flexibility earns its keep, consolidate where it does not, and be deeply suspicious of anyone selling an architecture as an identity. The next architectural movement, almost certainly built around AI agents, will arrive with the same confident framing. The useful habit MACH taught the industry, the hard way, is to ask what it actually costs to run before you believe the slide.
Council summary
This post argues that MACH was a sound architecture oversold as a mandatory one, and that VTEX's March 2025 exit from the MACH Alliance exposed the real bill: integration glue, compounding best-of-breed contracts, and operational friction for business users. The council verified the core facts against primary sources: the 26 March 2025 exit, the June 2020 founding by commercetools, Contentstack, EPAM, and Valtech, the October 2024 brand refresh, the September 2025 leadership change to Jason Cottrell, the 100-plus membership, and the February 2026 research showing 78 percent versus 13 percent AI ROI. Two misattributed cost figures were cut: a claimed three-year composable-versus-headless premium with no source, and a 20 to 40 percent maintenance figure that actually describes monolithic customisation, not composable builds. An invented "18 months" degradation timeline and an inflated founding membership count were also corrected. The takeaway for a reader planning a commerce build is to price the integration headcount up front, decouple only where flexibility earns its keep, and treat any vendor coalition's architecture as an input rather than an instruction.
Comments