After more than a decade of building and running SFCC storefronts, we kept seeing the same pattern. Good ideas don’t stall because teams can’t write the code. They stall in the distance between the merchant who needs a change and the developer who builds it: handoffs that thin out context, refinement and estimation cycles that take weeks, and quotes built from meeting notes rather than a deep read of the repository.
AI changed the shape of that problem. It collapsed the gap between intent and working code, which means the slow part of delivery has moved. Estimation and planning made sense when writing code was the bottleneck. It isn’t anymore.
Tokenwright is built for that new shape. AI gives our operators instant, deep context across your entire codebase — depth that used to take months to accumulate. Senior people handle the judgment; the engine handles the volume.
Every task follows the same four stages, from a one-line fix to a multi-week feature:
A Managed Token is a small, fixed unit of delivered SFCC work — and the unit you spend. A small feature or fix typically lands between 80 and 150 tokens; a mid-size feature between 300 and 700. Pricing reflects the value of what ships — a merge-ready pull request — plus the senior review and the liability behind it, never a meter of hours or compute.
The cap on every quote is firm. If the work runs over, the overrun is on us.
Once a codebase is onboarded, the engine keeps running. The Q&A Agent lets merchandisers, product managers and IT leads ask plain-English questions about how the store actually works — “Why don’t loyalty points apply to sale items?” — and get answers grounded in the real code. The Monitor Agent watches sandbox logs and the repository for errors, anomalies and security risks, and surfaces each one with a quote-ready fix. Both are included with every account, and more agents are on the roadmap.
Tokenwright doesn’t replace your system integrator. It sits on top as a velocity layer, clearing the backlog while your existing partner keeps production, monitoring and pager duty. Access is read-only — a sandbox or read-only repository — and production environments, customer data and secrets are never touched.
When a client wants more, ZaUtre is underneath: the same senior SFCC team that has taken over and run platforms for brands like Montblanc, Samsoe Samsoe and Multiopticas, ready to absorb deploys, support and 24/7 operations. That’s a conversation clients raise, not one we pitch.
Tokenwright is in private launch, and admissions are capped to protect senior-operator capacity. Signing up takes a minute — no SOW, no sales call — and every admitted account starts free with the Q&A Agent, scoping and 300 starter tokens.
]]>Recommerce — the resale of pre-owned goods — is growing faster than new retail in almost every product category. For retailers, it represents an opportunity: a way to recapture revenue from the secondary market, deepen customer loyalty, and meet growing consumer demand for sustainability.
The operational challenge is significant, though. Unlike new products, pre-owned items are each unique. Every item needs individual assessment: condition grading, pricing, authentication, photography. At any meaningful volume, doing this manually is either slow, expensive, or both.
PICAsell addresses this through an AI-assisted intake workflow that reduces the time and expertise required to list a pre-owned item from an average of 20 minutes to under 3.
The core of PICAsell’s AI capability is a multi-step product intelligence pipeline:
1. Visual condition assessment. A retailer photographs the item using a guided capture flow — the app instructs them to capture specific angles. The images are analysed by a computer vision model that identifies wear, damage, and condition indicators. The output is a standardised condition grade (New, Like New, Good, Fair) with supporting evidence surfaced for human review.
2. Automated description generation. The condition assessment, combined with product catalogue data (brand, model, materials), feeds into a language model that generates a product description optimised for discoverability and trust. Descriptions are brand-voice aware — a luxury resale listing reads differently than a sporting goods listing.
3. Dynamic pricing. Market price signals from comparable listings are combined with condition grade and demand data to generate a recommended price. The system learns from sale outcomes over time: items that sell quickly relative to their recommended price inform future recommendations.
4. Provenance and authentication. For high-value categories — luxury goods, limited-edition sportswear, collectibles — PICAsell uses blockchain to issue a verifiable provenance record at the point of intake. When the item sells, the record transfers to the new owner. This addresses one of the primary barriers to recommerce adoption in the luxury segment: the buyer’s inability to verify authenticity independently.
Building PICAsell has refined how we think about AI integration in commerce products generally.
Trust calibration matters more than accuracy. A 90% accurate condition grading model is not useful if retailers don’t understand when to trust it and when to override it. We spent more time on the explainability layer — surfacing why the model made a particular assessment — than on pushing the model’s accuracy from 90% to 92%. The explainability work drove adoption. The accuracy improvements on their own did not.
The workflow design is the product. AI handles the repetitive, pattern-matching parts of product intake. But the workflow around it — how retailers are guided through photography, how they review AI outputs, how exceptions are escalated — determines whether the system actually gets used. We iterated on the intake UX far more than we expected to.
Integration complexity is the real bottleneck for retailers. The technical capability to automate intake is not the constraint. The constraint is integrating a recommerce workflow into existing retail operations: connecting to the retailer’s PIM, syncing inventory states, routing items through existing warehouse flows. This is where a commerce-native team — one that has built and maintained these integrations for primary commerce — has a real advantage.
PICAsell is in beta with a small number of retail partners across fashion and sporting goods. We’re expanding the beta later this year, with a particular focus on luxury accessories — a category where authentication and provenance are highest-value.
If you’re a retailer exploring recommerce and want to understand what an AI-assisted intake platform could mean for your operation, we’d be glad to have a conversation. Reach out here.
]]>The term has been stretched to cover a wide range of things, so it helps to be precise. A genuinely composable architecture:
This is different from “headless SFCC,” which is still a monolithic commerce engine with a decoupled frontend. Both are valid approaches, but they are not the same architecture.
Luxury brands have specific characteristics that change how composable architecture should be evaluated.
Conversion rate is not the primary metric. For a luxury brand, a purchase is often the outcome of an extended relationship — clienteling, in-store touchpoints, event invitations, editorial content. The e-commerce site is one node in a broader brand experience, not the primary conversion channel. This means the ROI model for architectural decisions looks different than it does for a pure-play e-commerce retailer.
Brand consistency across markets is critical and hard. Luxury brands operate across many markets with different regulations, currencies, tax structures, and fulfilment partners. Composable architecture’s promise of mixing best-of-breed services per region sounds appealing, but in practice it multiplies the integration surface and the number of vendor relationships to manage.
The frontend is the product. For a luxury brand, the quality of the digital experience is brand-defining. Composable architecture gives you complete control over the frontend, which is genuinely valuable — but it also means your frontend is entirely your responsibility. You cannot lean on a platform’s reference design.
The strongest case for moving toward composable architecture is when a brand has outgrown what a monolithic platform can do and has the in-house or external technical capacity to manage the resulting complexity.
Specific triggers we see:
For brands that are primarily focused on transactional e-commerce — particularly Salesforce Commerce Cloud implementations — the platform continues to provide substantial value through native integrations, an active ecosystem of cartridges, and a well-understood operational model.
A mature SFCC implementation with a well-structured cartridge architecture and a disciplined approach to customisation can serve a luxury brand well for years, and does so with far lower operational overhead than a fully composable stack.
The honest answer to “should we go composable?” is almost always: “it depends on where you are in your digital maturity, what your roadmap looks like, and what engineering capacity you have or can build.”
Start with an architectural assessment rather than a platform decision. Map your current pain points against what your architecture can and cannot address. The answer often reveals that specific capabilities are the constraint — not the overall architecture.
We’ve helped several brands work through this decision, including some that chose to stay on SFCC and invest in composable-adjacent patterns (headless frontend, API-first integrations) rather than a full platform migration. It’s a more nuanced space than the vendor marketing suggests, and the right answer is genuinely different for each brand.
If you’re navigating this decision, we’d be glad to talk through your specific situation. Get in touch.
]]>Salesforce deprecated Site Genesis in favour of SFRA (Storefront Reference Architecture) years ago, yet a significant portion of active SFCC storefronts still run on it. The reason is straightforward: the migration has no obvious forcing function until something breaks. Maintenance costs creep up, community cartridge support thins out, and eventually a compliance requirement or a platform upgrade forces the decision. By that point, the technical debt has compounded.
Most platforms have 20–40 custom or third-party cartridges accumulated over years. Each one needs to be audited: Is it still used? Is there an SFRA-compatible version? Does the vendor still support it? We’ve seen cartridges that are called in exactly one business flow, undocumented, that no one remembered adding.
Plan for at least two weeks of cartridge inventory work before writing a single line of migration code. The findings from this phase determine your entire timeline.
Site Genesis controllers accumulate business logic over years of quick fixes. SFRA moves to a pipeline model that encourages cleaner separation, but the migration doesn’t automatically untangle that logic — someone has to read it, understand it, and consciously decide where it goes. For a mid-sized storefront, this review alone can take a month.
Checkout is where the most business-critical customisations live: loyalty integrations, payment providers, address validation, gift options, B2B quote flows. These rarely have comprehensive test coverage in Site Genesis implementations, because they were often built under deadline pressure. Before migration, write end-to-end tests against your existing checkout. They become your regression baseline.
Teams often staff the QA phase too lean, expecting that a side-by-side comparison is sufficient. It isn’t. SFRA renders differently, handles session state differently, and has different caching behaviours. Plan for a dedicated QA phase of at least three weeks for a mid-complexity store, longer if you have internationalisation requirements.
Page Designer is often a pleasant surprise. Brands that migrate find that giving merchandising teams direct content control — without developer involvement for every banner change — pays back the migration cost in the first six months.
Performance is the other consistent win. SFRA’s cleaner architecture is easier to optimise, and the move to SFRA is often the trigger for a broader performance audit that results in meaningfully faster page loads.
We run migrations in three phases:
The total timeline for a mid-complexity storefront — one with a custom checkout, several integrations, and a decade of accumulated cartridges — is typically 16–24 weeks when done properly. Attempts to compress this below 12 weeks almost always result in a deferred second project to fix the first one.
If your team is approaching this decision, we’re happy to do a cartridge inventory assessment as a starting point. Reach out through our contact page.
]]>With this funding we are building PICAsell — a platform that empowers fashion and luxury retailers to enter the fast-growing second-hand market. PICAsell combines AI-assisted selling, blockchain-based product authentication, dynamic auctions, and sustainability tracking into a single solution, enabling retailers to unlock new revenue streams from pre-owned goods while building customer trust and demonstrating measurable environmental impact.
If you are interested in learning more about PICAsell or exploring how reCommerce can benefit your business, get in touch.
This project is co-funded by the European Union through the European Regional Development Fund under the “Research, Innovation and Digitisation for Smart Transformation” Programme 2021–2027.
]]>