The Brew
Ha Ha
Zero results is a signal now
The Webscale AI × Webeyez partnership is live
A shopper types “caliper 2018 silverado 1500 rear” into the search bar. The catalog returns nothing. At eight times normal load, they’re gone in seconds.
7 min read
Also in this issue
- Four new base images before the freeze window
- TackDriver Assist goes live on Ravin
- Eight-step BFCM pre-flight
Contents
Seven departments, one partnership launch and a pre-flight checklist for BFCM.
- 01The first pourThe friction that only shows up at 8x trafficp. 01
- 02Order upShip log, every item traced to a dated pagep. 04
- 03Under the hoodThe signal path, the missing webhook and why the image catalog is moving fastp. 06
- 04Bean counterTwo figures on the cost of invisible frictionp. 08
- 05Field notesTackDriver Assist goes live on Ravin Crossbowsp. 09
- 06From the blogHow to stop losing sales during traffic spikesp. 10
- 07The grindBFCM pre-flight checklist, eight stepsp. 11
The friction that only shows up at 8x traffic
A shopper types “caliper 2018 silverado 1500 rear” into the search bar. The catalog returns nothing. On a slow Tuesday they might try again. At eight times normal load, they’re gone in seconds and the storefront never gets the chance to recover the sale.
A new partnership with Webeyez closes that gap. Webeyez reads every live session in the background and names the break the moment it happens. When the signal fires, the AI Shopping Assistant opens on the same page, in the same session, and points the shopper to a real product. Signal to reply runs under two seconds end to end.
The integration went generally available on Aug 14. It’s live on Adobe Commerce and Magento, Shopware and Shopify. No storefront redesign, no new stack.
Named triggers ship today
For a developer, that’s a signal-triggered recovery layer with named events wired into the assistant already running on your site. Named triggers ship today: search_no_results, page_latency_high, unproductive_browse, exit_intent, idle_on_fitment and a custom_trigger built to your storefront during onboarding.
The whole path runs inside one session.
What this does not do
There’s no generic outbound webhook yet. If you want to correlate friction events with infrastructure telemetry in your own tooling, that capability is on the roadmap. For now the signal path lives inside the assistant.
Signal to reply, under two seconds.
Order up
Ship log
Every item traces to a dated page on docs.webscale.com.
Partnership
Webscale AI × Webeyez is GA
Webeyez signals now trigger the AI Shopping Assistant inside the same shopper session. Available on Adobe Commerce and Magento, Shopware and Shopify. Announced Aug 14.
CodeDirect
Builders reference refreshed
The docs for how CodeDirect assembles container images got a pass on Aug 21. If you’re pinning versions or writing custom build steps before the freeze, diff against this page.
Image catalog
2026.37 through 2026.40 shipped Aug 21
Four release tags in one day. If you build against a moving tag, pin now. Freeze windows come earlier than you think.
Support
Service Matrix updated
Aug 18. The response-time and coverage details changed. Read it if you’re setting expectations for a new deploy or renegotiating scope going into peak.
Reference
Software Products page refreshed
Aug 18. The canonical list of what’s included per plan. Useful if a customer or a sales conversation asks what’s in the box.
Hosting V3
Configuration docs refreshed
Aug 11. V4 gets most of the attention right now, but V3 environments still power a lot of production. If you’re on V3 and staying there through BFCM, re-read this.
docs.webscale.com/latest-updates →The signal path, the missing webhook and why the image catalog is moving fast
Notes on the plumbing, starting with how the Webeyez integration works under the hood.
Signal path
End to end, inside one session
The Webeyez integration runs client-side. Their script watches session state in the background and, when a break matches one of the named signals, calls the assistant through a JavaScript API with the context attached. The payload includes the trigger name and any relevant session state. For search_no_results that’s the failed query itself.
Latency numbers come off the partner page. Webeyez fires the signal in about 0.4 seconds. The assistant reply lands around 1.2 seconds. The full loop is under two seconds, and it happens without a page reload.
That last part matters more than the numbers. Because the assistant opens in-session, session state stays intact. Cart contents, cookies, applied promotions and browsing context all survive.
What’s not shipped yet
The missing webhook
Webeyez doesn’t emit a generic outbound webhook today. Signal routing lives inside the integration. If you’re trying to correlate friction events against infrastructure telemetry in your own stack, or feed them into a data pipeline, that capability is on the roadmap. Track it if control-panel correlation matters to your architecture.
Image catalog
Four base images in one week
Four base images landed on Aug 21. In a normal month you’d see one, maybe two. The pattern tells you the base is moving faster than usual right now, which usually means CVE churn and dependency updates ahead of the freeze.
If you build against a moving tag, pin it. If you already pin, diff the tag you’re on against 2026.40 before your freeze window opens. The gap between what’s in production and what’s in the current image is the gap you inherit if you have to roll a security patch mid-peak.
Bean counter
Two figures on the cost of the friction you can’t see
Two figures / Tier 1 sourcing
$234B
Lost by US retailers every year to search abandonment. Most of it invisible until the quarter closes.
Google Cloud and Harris Poll, 2023. Retailer survey. Tier 1.
2%
Conversion drop for every additional second of page load time on a large ecommerce site. The relationship holds across mid-market commerce.
Walmart engineering benchmarks, referenced in Adrian Luna, “The cost of a one-second page load delay,” webscale.com, July 21, 2026. Tier 1.
Field notes
TackDriver Assist goes live on Ravin Crossbows
Ravin Crossbows put the AI Shopping Assistant into production on Aug 13, running under their own brand as TackDriver Assist. Merchant-branded name on the storefront, standard assistant underneath.
A week in, it’s serving live shopper traffic and the tuning loop is running. When live feedback flagged a wrong product recommendation, the fix went in as a per-merchant custom prompt instruction. No deploy required. That’s the feature Arvind shipped on May 22 doing its work three months in.
See it in action →
How to stop losing sales during traffic spikes
By the time your server notices the surge, the sale is already lost. Adrian’s take on scaling from commerce signals instead of CPU load. Read this before your peak-load simulation gets scheduled.
The grind
BFCM pre-flight
Work through this in one sitting.
- 01Confirm which Hosting V4 tier each environment runs today
- 02Rebuild your base image against the latest 2026.4x tag
- 03Baseline your current failed-search rate and set the alert threshold now
- 04Audit VCL against the Vinyl 9.x breaking changes if you haven’t already
- 05Diff cache parameters against the V4 documented defaults
- 06Pull ASN distribution from traffic logs and add Web Controls conditions for the sources you want to shape
- 07Confirm the AI Shopping Assistant is deployed and Webeyez signals fire in the same session in staging
- 08Run a peak-load simulation while you still have staging capacity
Every two weeks
Forward this to whoever owns the deploy
The platform changes, and what to do about them in your environment.
Subscribe