Business
5 min readThe composable commerce debate used to be all architecture talk. Headless or monolithic? Microservices or all-in-one? We spent years arguing about what to build.
Now the conversation's different. It's about how you build and who gets final say when your business needs something the platform wasn't designed for.
What changed? AI agents that need direct system access to function. Customer expectations that move faster than vendor roadmaps. And companies that disagree completely on whether they want to run their own infrastructure. You can't solve any of this if you don't control your stack.
GenAI browser traffic to retail sites went up 4,700% last year. Over half of consumers expect to shop using AI assistants by year-end.
An AI agent connects to your catalog, checks pricing and inventory in real time, applies discounts and loyalty points, handles subscription preferences, and completes checkout. The whole transaction happens in conversation.
Integrating these tools requires specific access points. APIs that expose product data in structured formats. Hooks for real-time price negotiation. Order management that hands off to agents properly. Payment processing with fraud detection that works when a bot's making the purchase. OpenAI built a protocol for this. Google and Shopify made one together. PayPal launched infrastructure specifically for agents.
The problem is most platforms weren't built with this in mind. If you can't extend your APIs to expose data the way these agent tools need it, or modify checkout logic to handle agent-initiated transactions, you're waiting for your vendor to build it. When the technology's moving this fast, that doesn't work.
Product configuration is standard now. Steelcase customers build office layouts by dragging desks and chairs around. The system validates compatibility and updates pricing live. Vaillant does the same for heating systems, checking property specs and local regulations.
When you bolt it on through separate integrations, it breaks every time something updates.
Multi-region B2B gets complicated fast. You're managing catalogs where certain products only show for specific accounts. Pricing comes from contracts, not price lists. Approval workflows route through your ERP with different rules per business unit. Tax and compliance logic shifts whenever regulations change.
Subscriptions have their own issues. Usage billing. Upgrade paths that make sense for your product. Prorated charges. Overage fees. Renewal rules that vary by customer type. If you can't touch the billing engine, you end up building workarounds that create debt.
Every platform makes assumptions about how you operate. When those assumptions are wrong, you need to change the code.
What you're allowed to do with platform code depends on the license. Companies have been moving away from pure open source for good reason.
Open source means anyone can take the code and do anything with it, including compete against you. AWS can grab your software and sell it as a service while undercutting you on price. MongoDB, Elasticsearch, and Redis all ran into this and changed their licenses.
Source-available gives you code access with limits, usually around reselling it as a competing service. You can still read the source, modify it, deploy it wherever, and fork it if needed. You just can't build a rival-managed platform with it.
What this means when you're buying a platform: you get customization without vendor lock-in, and the vendor's protected enough to keep investing in the product. You're not depending on community contributors or hoping the business stays solvent.
You can fork individual services. Extend APIs for your integrations. Adjust data models for your catalog structure. Your changes sit alongside the core code instead of replacing it, so you can still take updates.
Proprietary platforms don't give you any code access. Pure open source might leave you stranded if the company behind it folds.
Having access to code doesn't mean you run everything yourself.
Our framework is source available for our licensed clients. You can host it in your cloud, our cloud, or split workloads; however, compliance requires it. Fork services when business logic demands it. Extend APIs when you need custom integration points. Modify data models when your catalog's different.
Or go fully managed. We handle scaling, patching, performance, databases, and monitoring. Same code, different operational setup. Pick what works for your team.
AI agents are early. Right now, it's recommendations and conversational checkout. Next, it's automated procurement. Contract negotiation. Multi-system transactions.
Task capability for language models doubles every seven months. In 2019, they could handle a few seconds of work. Now they complete 30+ hours of skilled labor.
Your platform has to keep pace without you waiting on vendor release schedules. You need code access when integration requirements change, when new data formats emerge. When what differentiates your business isn't available as a standard feature.
One-size-fits-all doesn't cut it anymore. Commerce is too complex, and companies are too different in how they operate and what they're willing to manage themselves.