← Journal
Author
Szymon Żynda
Co-founder of Seedlight · eCommerce platforms, AI, SEO and GEO
YouTube: Daniel & Simon | Commerce & AISzymon Żynda
Szymon is a co-founder of Seedlight, the technology brand of the Amazonway group, building, automating and growing eCommerce platforms on the BEAM framework: from Blueprint and engineering, through practical AI automation (product descriptions, translations, feeds, marketplace content), to maintenance and growth work.
He combines engineering and eCommerce builds with AI, marketing, SEO and GEO: making sure the platform not only runs, but the brand stays visible in Google and in AI answers. He co-hosts the YouTube channel "Daniel & Simon | Commerce & AI Visibility".
Areas of expertise
01
eCommerce platform builds and migrations
02
AI automation in commerce (the BEAM framework)
03
Integrations, product data and operations
04
Marketing, SEO and content
05
AI Visibility (GEO): brand presence in AI answers
Articles by this author
AI EngineeringClaude can drive a computer and a browser: what shipped, and what it means for a storeComputer use left beta on 19 August 2026, and a separate browser tool launched the same day. Two different things share that name, though, and only one of them touches your machine. The specification, the costs, and the risk the documentation names outright.AI EngineeringClaude Fable 5.1 and Mythos 5.1: specs, benchmarks, and what changes for usAnthropic released two models on 1 September 2026. Full specifications, every published benchmark, a comparison against the rest of the Claude family, the new pricing, and an honest answer on what it changes in how Seedlight works.AI Visibility42 percent of our AI crawl was not AI. How to measure yoursFor a month we believed AI models had fetched our content 6,600 times. After filtering, 3,934 remained. Three measurement mistakes that inflate AI traffic figures almost everywhere, and how to check your own logs.AI VisibilityWe checked 24 Polish stores: can an AI agent buy from them?We crawled product pages across twenty-four Polish stores and checked nine fields a shopping agent needs to complete a purchase. Three of them appeared nowhere, and seventeen stores could not be read at all. Method, limitations and the raw numbers.eCommerceAI product recommendations: how much data you need before they earn anythingRecommendations learn from interactions per item, not from total orders. A calculation you can run on your own numbers, four approaches with the data threshold for each, and why measuring without a control group always reports success.AutomationsAI assistants for online stores: what works today and what is still too earlyOne label covers two different products: the advisor your customer talks to, and the assistant working in your back office. Buyers ask about the first, the second pays back sooner. Why that is, how to test your catalogue in an hour, and the three cases where an advisor earns from day one.AI VisibilityWhat an AI agent sees when it comes to buy in your storeAn agent never looks at your page, it receives a record. Field by field: what has to be in that record before a purchase can close, three silent failures worse than a missing field, and the standards decision (ACP, UCP, the Shopify agent flow) that nobody will make for you.B2BB2B ordering portal: what it must do before your reps stop retyping emailsA portal only empties the sales inbox once it closes a whole order: this buyer’s price, a reorder from history, a pasted SKU list, the credit limit, real availability dates, approval on the buyer side, a PO reference and an honest status when the ERP goes quiet. Each requirement with the condition it fails without, plus an acceptance test to run on a demo.eCommerceWho should build your eCommerce platform: a freelancer, an agency, a software house or an in-house teamFour delivery models held to the same standard: when each one is right, how it bills, what is left when it goes away and where it usually breaks. With a decision table, the questions to ask before signing, and the things builders do not publish about themselves. Checked 17 August 2026.AI EngineeringHow an agency works with Claude Code on a client’s storeAn AI agent writing code on your platform is a trust question, not a productivity question. What agents are never allowed to touch, how a change reaches production, how you approve it without reading pull requests, and what stays yours when the contract ends.Global ExpansionMedusa multi-language: how translations work and what breaks at scaleSince version 2.12.3 Medusa stores translations natively, which settles where multilingual content lives and none of what actually breaks in bulk translation: markup, brand names, units and field limits. Here is what the Translation Module covers as of 17 August 2026, the six failures that repeat, and the checks that catch them before publication.B2BFramework agreements in German B2B: what a Rahmenvertrag settlesGerman wholesale buyers start with an agreement, not an order. What a Rahmenvertrag usually settles, which billing models follow from it, where German law fixes payment periods and late payment interest, and what your platform has to carry so the terms survive contact with real orders.eCommerceWhat an eCommerce build costs on every platform, and who publishes the numberPublic implementation ranges for eight eCommerce platforms, and next to every figure the column nobody else prints: who published it. Including a section on the numbers you should not trust. Checked 13 August 2026.AutomationsAI agents running your store: what they do on Monday morningAn AI agent in a store is really two things: a workflow triggered by an event, and an agent answering questions about live data. Concretely: what happens when 200 products land, when a feed rejects 30 items, and when a customer asks where the parcel is. Plus the permission model we insist on before any AI touches an order panel.eCommerceBEAM + Medusa vs Shopify Plus and an agency: two routes, not two productsA comparison of offers rather than engines: what you get on day one, who runs it, whose platform it is and what you keep if you part ways. With a three-year model built from substitutable assumptions, and our published price ranges.eCommerceShopify alternatives for growing brands: seven routes and one testSeven realistic directions for a brand that has outgrown Shopify, six of which we do not sell. Each one with two numbers, platform cost and build cost, because every option on this list has to be implemented by someone. With an honest section on when to stay.eCommerceMedusa vs WooCommerce: free to start, paid to maintainBoth licences cost nothing. The difference sits in the build, the running cost and the price of the first change after launch, so we split the bill into three explicit layers. Twelve dimensions side by side, plus an honest section on when WooCommerce is the cheaper answer. Vendor data checked 11 August, build ranges 12 August 2026.eCommerceMedusa vs Shopify: which one fits, and when switching is a mistakeMedusa and Shopify compared across eleven dimensions: licence, cost to implement, cost to change after launch, checkout, API limits and lock-in. Figures checked at source, current as of 12 August 2026.eCommerceBlack Friday: an engineering checklist for your platformBlack Friday 2026 falls on 27 November. An engineering schedule: what to measure in September, what to rebuild in October, what to freeze in November.ScalingYour eCommerce team after launch: who you actually needAfter launch you need four functions covered, not four new hires. Which roles scale with order volume, and which are better bought than employed.AI EngineeringThe agent did exactly what we wrote: why AI automations fail quietlyAI automation rarely fails loudly. It follows the instruction literally rather than the intent, so the same defect lands in every single output. Three measured cases from our own work, applied to descriptions, attributes and feeds.AutomationsAI catalog translation without an agency: where translation ends and localization beginsBulk catalog translation is one of the best first AI projects in a store: it runs on data you already own, and you catch every mistake before it goes live. On one condition, that you know where translation ends and localization begins, because that is where most failed rollouts break.Global ExpansionEntering the German market: an eCommerce checklistTranslating your store is the smallest part of selling in Germany. The rest is packaging registration, VAT and OSS, the Impressum, product data in the format German channels expect, payment methods you may not offer yet, and returns run as a process. Here is the checklist in the order that works.AI EngineeringeCommerce skills for AI agents: why we open-sourced oursWe have published a set of AI agent skills specialised in eCommerce: catalog and feeds, structured data and AI visibility, 2026 compliance, migration and a pre-deploy gate. What is inside, how to install it, and why the safety rules in a skill matter more than the list of steps.AutomationsAI for order support: automating „where is my order" (WISMO) questions„Where is my order" (WISMO) is the most common and most repeatable category of contacts in eCommerce. How eCommerce customer service automation takes it over in a draft-first model: AI writes a draft with order context and status, and a human approves or escalates.eCommercePOAS, not ROAS: measure profit from ads, not revenueROAS shows the revenue an ad touched, not the result. Two campaigns with identical ROAS can leave one in profit and the other in loss, because margin differs. POAS (Profit on Ad Spend) measures what actually stays: definition, formula, a worked example and the scaling trap.AI VisibilityHow AI describes your brand: Brand Visibility as a new KPIHow ChatGPT, Perplexity and Gemini describe and recommend your brand is becoming a metric of its own, because customers increasingly ask an AI assistant instead of Google. How to check for yourself what a model says about your brand, what shapes that picture, and what you can actually fix, without promising the AI will start recommending you.AI EngineeringMarketplace feeds with Claude Code: from export to validationSame catalog, but every channel wants its feed its own way: different field names, different formats, its own taxonomy and hard rejection rules. Here is how Claude Code maps catalog columns onto a channel's fields, fills gaps, validates types and values, and flags offers that will be rejected before you upload. On a copy of the data and with a diff review.AI VisibilityZero-click commerce: the end of the click and what it means for your storeMore and more often the customer gets an answer to a buying question right away, in the results or from an AI assistant, and never lands on your site. What that means for your store, why it is not the end of the world, and how to become the source of the answer the model shows the buyer instead of losing traffic.AI VisibilityAgentic commerce: how to prepare your store so AI recommends itAgentic commerce means AI agents shop on the buyer behalf: they search, compare and increasingly close the purchase. You prepare your store for this not with a campaign, but by making it readable to a machine. A concrete checklist, plus how to check whether AI already recommends you.AI VisibilityGEO for eCommerce: how to land in ChatGPT and Perplexity answersGEO is not a separate kind of magic detached from SEO. It is the same good SEO, extended to visibility in generative answers. What to actually do so your brand and products show up in ChatGPT and Perplexity, and how to measure that presence, without promising miracles.AI VisibilityStructured data for AI: the Product, Offer and FAQ agents look forFor search engines and AI agents to understand your store unambiguously, you need a handful of schema.org types in JSON-LD: Product, Offer, AggregateRating and Review on real data, BreadcrumbList, FAQPage and Organization. Which ones matter, how to implement them correctly, and the mistakes that draw a penalty instead of visibility.AI EngineeringClaude Code for eCommerce: cleaning a catalog CSV step by stepA messy catalog CSV export can be cleaned in a few passes: split variants, validate EAN/GTIN codes, normalize units and colors, repair encoding, generate slugs, flag duplicates and gaps. Here is the full process and how to brief it, on a copy of the data and with a diff review before you save.AI EngineeringAgentic engineering: how we build ecommerce platforms with AI agentsAn AI agent writes code fast, but the bottleneck is verification, not writing. We show the process where gates, an adversarial review and a human on irreversible actions deliver speed and safety at once.AutomationseCommerce automation: where to start and how many hours it really savesWhere do you start automating an online store, and how much time does it genuinely take off your team? Instead of one magic number: a qualification rule, seven repeatable tasks with hours before and after, and the work that always stays with a person.B2BB2B contract pricing: the layers of price and which rule winsB2B pricing is not one discount but several models running at once: price levels, customer groups, individual prices, volume tiers and framework agreements that fix terms before the first order is placed. Here is the order in which the rules resolve, what a framework agreement has to settle, and what your platform and ERP need to carry.ScalingWhat Breaks at 10x Traffic: The Engineering of Scaling an eCommerce StoreWhen traffic suddenly spikes, a store rarely fails for a single reason. The database gives out first, then the missing cache, and finally the synchronous integrations. Here is what breaks as traffic grows and how to prepare your platform for the peak before it arrives.eCommerceWhich ecommerce platform? Criteria, not rankingsThere is no best ecommerce platform, only one that fits your business. Instead of a ranking: nine criteria that actually decide your migration target, plus an honest when SaaS is enough.eCommerceThe checkout that stops losing carts: a 10-point auditOver 70% of carts end up abandoned, but the recoverable ones have repeatable, concrete causes. A 10-point checkout audit: what to check, the typical mistake and how to fix it.eCommerce12 signs your store has outgrown its SaaS platformTwelve signs a SaaS platform is limiting your store, what each one really costs, and a simple framework for when to migrate and when to stay.B2BWholesale digitisation in 90 days: a real plan, not a mega-projectDigitising wholesale does not have to be a year-long project with an uncertain end. Here is a 90-day plan in three phases where each ends with something that works, not a slide deck.eCommerceThe true cost of owning your eCommerce platform: a 3-year TCOSaaS looks cheap in month one and expensive in year three. We break down the total cost of ownership: subscriptions, fees and apps versus building a platform you own.B2BB2B store and ERP integration: an architecture that does not hurtThe most expensive word in wholesale is retyping. Here is the store-to-ERP architecture, sync rhythms and error handling that make orders flow on their own.AutomationsAI product descriptions done right (without hurting SEO)A description generator is not a chat window. Here is the workflow that holds brand voice across thousands of SKU, passes human review and meets Google guidelines.AI EngineeringAI-native is not vibe codingWe build every project with Claude Code. But AI without engineering discipline is chaos in production. Here is how we combine both.eCommerceYou are not buying hours: why fixed scope winsHourly billing shifts all the risk to the client. In BEAM every stage has a fixed price and a defined outcome. Here is why we price this way.eCommerceMigrating a store without stopping salesThe biggest fear of replatforming: "what about sales during the move?". Our migration playbook answers with specifics, not reassurances.
Building or growing an eCommerce platform?
A 30-minute call with the founders. We come prepared with an initial read of your situation.
