DEV Community: klyx

Setting Up Wholesale Pricing on Shopify Without a Separate Store

klyx — Sat, 14 Nov 2026 14:39:00 +0000

Setting Up Wholesale Pricing on Shopify Without a Separate Store

Merchants often assume wholesale means standing up an entirely separate store. Shopify's native B2B features (and several apps) handle tiered, customer-specific pricing on the same storefront without that overhead.

🏢 Shopify's Native B2B Feature

Available on higher Shopify plans, this lets you create company profiles with custom price lists, payment terms, and minimum order quantities — all on the same storefront, visible only to logged-in wholesale accounts.

🏷️ Tag-Based Pricing for Simpler Needs

For a simpler setup, customer-tag-based discount apps can apply a flat wholesale discount percentage to tagged accounts without the full B2B feature set — less flexible, but faster to set up for straightforward tiered pricing.

🚧 When a Separate Store Still Makes Sense

If wholesale needs a genuinely different catalog, completely different branding, or checkout flows too divergent from retail, a separate store is still the right call — but that's a higher bar than most merchants assume going in.

Originally published on the KLYX blog.

Managing Context Windows for Long-Running Agent Tasks

klyx — Sat, 14 Nov 2026 11:22:01 +0000

Managing Context Windows for Long-Running Agent Tasks

A task that runs long enough eventually fills whatever context window it has, and what happens next determines whether the agent stays coherent or starts losing track of earlier decisions.

📝 Rolling Summarization

Periodically compressing older turns into a dense summary while keeping recent turns verbatim preserves the gist of earlier decisions without paying full token cost for them indefinitely — the standard approach for conversational tasks.

💾 Externalizing State Instead of Carrying It

For tasks tracking structured progress (a checklist, a set of decisions made), writing that state to a file or memory store the agent can re-read is more reliable than hoping it survives compaction inside the context itself.

⚠️ Where This Still Breaks Down

Tasks requiring precise recall of something said many turns ago, in exact wording, don't survive summarization well — if exact fidelity matters, that specific detail needs to be externalized explicitly, not left to compaction to preserve.

Originally published on the KLYX blog.

Do Product Videos Actually Convert Better Than Images? What We've Seen

klyx — Sat, 14 Nov 2026 07:07:00 +0000

Do Product Videos Actually Convert Better Than Images? What We've Seen

"Add video, boost conversion" gets repeated as a near-universal rule. In practice it depends heavily on the product category and how the video is actually implemented, not just whether one exists.

✅ Where Video Genuinely Helps

Products where functionality, fit, or scale are hard to convey in a still image — how a tool actually operates, how fabric moves, how large something is relative to a person — get real conversion lift from video, because it answers a question a photo can't.

➖ Where It Barely Moves the Needle

For simple, visually self-explanatory products, video adds page weight and production cost without meaningfully changing the buy decision — the static image was already answering the question the customer had.

⚠️ Autoplay and Load Time Can Backfire

A heavy autoplaying video that slows the page or starts with sound can cost more in page-speed and annoyance than the video gains in persuasion — click-to-play, lazy-loaded, is almost always the safer implementation.

Originally published on the KLYX blog.

Structured Output vs Function Calling: Picking the Right One

klyx — Sat, 14 Nov 2026 06:17:39 +0000

Structured Output vs Function Calling: Picking the Right One

Both constrain a model's output into something a program can reliably consume, and it's easy to reach for whichever one you used last time rather than the one that actually fits the task.

✅ When Structured Output Fits

If the goal is a single, well-defined data shape as the final answer — extracting fields from a document, classifying into fixed categories — structured output constraints are simpler and don't need the function-calling machinery around them.

✅ When Function Calling Fits

If the model needs to choose between multiple possible actions, potentially call several in sequence, or the task is genuinely agentic (decide what to do, then do it), function calling is built for that — structured output alone doesn't model "pick an action."

🔀 They're Not Mutually Exclusive

A function call's arguments are themselves structured output — the distinction is really about whether the model is choosing an action (function calling) or just shaping a final answer (structured output), and a real system often uses both at different points.

Originally published on the KLYX blog.

Hitting Shopify's Variant Limits: What Actually Works Around It

klyx — Fri, 13 Nov 2026 13:43:00 +0000

Hitting Shopify's Variant Limits: What Actually Works Around It

Shopify's variant limit catches merchants with highly configurable products off guard — usually right when a catalog is growing, which is the worst time to discover a structural constraint.

🔢 What the Limit Actually Is

Shopify caps variants per product at a fixed number based on option combinations — a product with too many option values multiplied together hits the ceiling fast (3 options with several values each compounds quickly).

✂️ Splitting Into Multiple Products

The most reliable workaround is splitting one over-configured product into several related products (grouped visually via a product list or collection), each staying under the limit — more admin overhead, but it's the stable, supported path.

🧩 App-Based Custom Options as an Alternative

For options that don't need separate inventory tracking (engraving text, a color swatch that doesn't change price or stock), a line-item properties app can add configurability without consuming the variant limit at all — worth checking whether the option genuinely needs to be a tracked variant first.

Originally published on the KLYX blog.

Scoping Tool Permissions for an AI Agent: The Pattern We Actually Use

klyx — Fri, 13 Nov 2026 10:11:00 +0000

Scoping Tool Permissions for an AI Agent: The Pattern We Actually Use

The fastest way to build an agent is to give it broad access to everything and let the prompt tell it what not to do. The fastest way to have a bad incident is the exact same decision.

🎯 Scope Access to the Specific Task

An agent that only ever needs to read product data and write to a specific spreadsheet shouldn't have write access to the full database, even if it's "never going to use it that way" — prompts are guidance, not a security boundary.

🔐 Separate Read Access From Write Access Explicitly

Most agent tasks need far more read access than write access. Granting broad read but narrow, explicit write permissions limits the blast radius of a mistake to the specific things the agent is actually supposed to change.

🚦 Irreversible Actions Get a Human Gate

Anything that can't be easily undone — sending an email, charging a card, deleting data — goes through a human confirmation step regardless of how confident the agent is. This is the one rule we don't make exceptions to.

Originally published on the KLYX blog.

Your 404 Page Is Costing You More Than You Think

klyx — Fri, 13 Nov 2026 06:27:59 +0000

Your 404 Page Is Costing You More Than You Think

A generic "Page Not Found" with nothing else on it is a dead end for both visitors and search engines — and most stores never audit how often visitors actually land there.

🔗 Why Visitors Land There More Than You'd Think

Discontinued products, old marketing links still circulating, and typos in URLs people type directly all funnel into 404s — checking Search Console's crawl errors report regularly surfaces exactly how much traffic is hitting dead ends.

🔄 Redirect the High-Value Ones

For discontinued products with real remaining search demand, a 301 redirect to the closest live replacement or category page recovers that traffic instead of losing it entirely — this is the single highest-leverage fix on most stores' 404 lists.

🧭 What a Good 404 Page Actually Does

Search bar, links to top categories, and maybe a discount nudge turn a dead end into a recovery opportunity instead of a bounce — treat it as a real page in the funnel, not an afterthought nobody's supposed to see.

Originally published on the KLYX blog.

Setting a Free Shipping Threshold: The Number Matters More Than the Feature

klyx — Fri, 13 Nov 2026 06:18:54 +0000

Setting a Free Shipping Threshold: The Number Matters More Than the Feature

Almost every store has a free shipping threshold at this point. The number chosen determines whether it's actually lifting order value or just giving away margin on orders that were already above it.

❌ The Threshold Set Too Low

If the threshold sits at or below your current average order value, you're giving away free shipping on orders that would have happened anyway — it's not changing behavior, just cutting margin on your typical order.

✅ The Threshold That Actually Nudges Behavior

Setting it 15-25% above current AOV gives customers a genuine reason to add one more item, which is where the feature actually earns its cost instead of just subsidizing orders that didn't need the nudge.

👁️ Visibility Matters as Much as the Number

A threshold that only shows up as a surprise at checkout doesn't influence behavior — a progress indicator visible in the cart ("$12 more for free shipping") is what actually drives the extra add, not the policy existing somewhere in the fine print.

Originally published on the KLYX blog.

Fixed Bundle or Volume Discount? How to Pick for Your Catalog

klyx — Fri, 13 Nov 2026 06:18:53 +0000

Fixed Bundle or Volume Discount? How to Pick for Your Catalog

These are the two most-used offer types in Bundler, and merchants often default to whichever one they heard about first rather than the one that actually fits their catalog.

📦 When Fixed Bundle Wins

If there's a specific combination of different products that genuinely makes sense together (a skincare routine, a starter kit), Fixed Bundle packages them at one price with zero decision burden on the customer.

📈 When Volume Discount Wins

For a single product or a tight product family where customers naturally buy more of the same thing (consumables, refills, multi-packs), Volume Discount rewards buying more without forcing an artificial bundle of different items.

🧭 The Quick Test

Ask: "does this offer need different products combined, or more of the same product?" Different products, different purposes → Fixed Bundle. Same or similar product, more quantity → Volume Discount.

Originally published on the KLYX blog.

Evaluating LLM Output Quality Without a Dedicated Eval Team

klyx — Thu, 12 Nov 2026 10:36:02 +0000

Evaluating LLM Output Quality Without a Dedicated Eval Team

"Build an eval pipeline" is right advice that assumes resources most small teams don't have. Here's a version of that discipline that takes an afternoon, not a sprint.

📋 Build a Small Golden Set, Not a Big One

Ten to twenty real examples covering the actual range of inputs the system will see — including the tricky edge cases, not just the easy ones — is enough to catch regressions. A huge eval set is nice eventually, not a prerequisite to start.

👤 A Manual Review Cadence Beats No Review

Reviewing a handful of real production outputs weekly, scored against a simple rubric (correct, acceptable, wrong), catches drift and prompt regressions long before a fully automated eval pipeline would even exist for a small team.

🤖 When Automating the Eval Actually Pays Off

Once volume is high enough that manual review can't keep pace, or a prompt change needs to be validated against the golden set before every deploy, that's the point automating pays for itself — not before.

Originally published on the KLYX blog.

How Shopify Discount Stacking Actually Works (and Where Merchants Get Surprised)

klyx — Thu, 12 Nov 2026 07:19:01 +0000

How Shopify Discount Stacking Actually Works (and Where Merchants Get Surprised)

Discount stacking behavior surprises merchants more than almost any other pricing feature — not because it's broken, but because the combination rules aren't obvious until you hit the exact scenario that breaks your assumption.

🏷️ The Three Combination Classes

Shopify groups discounts into product, order, and shipping classes, and by default only one discount per class applies at checkout — two product-level discounts don't automatically stack just because both are active.

🔗 Stacking Requires Explicit Setup

To let a product discount and an order discount apply together, that combination has to be explicitly configured when creating the discounts — it's opt-in, not automatic, which is exactly where merchants expect stacking and don't get it.

⚙️ Function-Based Discounts Add Another Layer

Discounts built with Shopify Functions (like bundle or volume discounts) follow the same class rules but are less visible in the admin UI when debugging a stacking question — worth testing the exact real-world cart scenario directly rather than assuming from the discount list alone.

Originally published on the KLYX blog.

Claude's Memory Tool vs Building Your Own RAG: What Small Teams Should Actually Pick

klyx — Thu, 12 Nov 2026 06:13:11 +0000

Claude's Memory Tool vs Building Your Own RAG: What Small Teams Should Actually Pick

Built-in memory features from model providers removed a lot of the reason to build custom RAG infrastructure for straightforward use cases. For a small team without dedicated infra time, that trade-off usually favors the built-in option more than people assume.

✅ When the Built-In Tool Is Clearly Enough

If the need is "remember facts and preferences across sessions for a single assistant use case," a provider's built-in memory tool does that out of the box, with no infrastructure to maintain and no vector database to keep in sync.

🏗️ When Custom RAG Still Earns Its Place

Once the knowledge base is large, shared across multiple products or teams, or needs retrieval logic specific to your data (custom ranking, access control per document), a built-in generic memory tool won't flex that far — that's genuinely a custom-build problem.

💸 The Real Cost Most Teams Underestimate

Custom RAG isn't just the initial build — it's ongoing maintenance: re-embedding when the model changes, monitoring retrieval quality, handling edge cases. For a small team, that ongoing cost is often the deciding factor, not the initial setup.

Originally published on the KLYX blog.