AI Commerce Weekly: Week 31, 2026
A quiet week with a clear shape. Discovery has moved to the agent, but only 7% of UK shoppers think the delivery promise is being kept. The genuine developments of the week were all at the loading dock.
TL;DR
W31 was quiet by count and clarifying by shape. Nobody shipped a new checkout rail or a shopper-facing flourish. The three developments that mattered all sat at the far, unglamorous end of the funnel, at fulfilment and returns. Vercel's Agent Era data, published for the UK on 30 July, says discovery has already moved to the agent: roughly a fifth of UK consumers now start a journey on AI search, and 43% say they would switch retailer for one that used AI better on its site. Then Locus put a number on the bill that discovery runs up. Only 7% of UK shoppers think fast-delivery promises are consistently met. And UNIT AI raised $12m to automate the fulfilment-and-returns half that everyone spent the quarter ignoring. The industry got very good at being found by an agent. It has barely started asking whether the warehouse can cope when the agent actually buys.
Discovery already moved to the agent
Start with the demand side, because it is the part that has actually changed. Vercel's "Agent Era" study, in the UK coverage that landed on 30 July, says the front of the journey has quietly shifted. About a fifth of UK consumers now begin shopping on AI search, ahead of Facebook, TikTok and Instagram, and 43% say they would choose one retailer over another because it makes better use of AI on its website. We gave that report a piece of its own last week, because the switching number reframes on-site AI as a churn lever rather than a nice-to-have. The short version for this desk: the shopper is arriving at your storefront having already asked an agent, and increasingly deciding whether to stay based on what the agent found.
What the same data refuses to say is that shoppers have handed over the buying. 79% still start on the retailer's own site, 65% still finish high-value purchases in a physical store, and only around one in ten would spend more with an AI assistant than with a human. It is worth being honest that this is Vercel's own commissioned research, so treat the exact figures with the usual caution. But the direction matches everything the feed has tracked all quarter, from CI&T to dunnhumby: people reach for an agent to find things, not to hand the wallet across. Discovery has moved. Delegation has not. And that split is precisely what makes this week's other two stories matter, because an agent that sends a shopper somewhere new is only as good as what happens after the click.
The promise the warehouse can't keep
Here is the number that should stop a Head of Technology mid-scroll. New Locus research, published on 30 July, surveyed 2,000 UK shoppers and found that agentic shopping is genuinely growing, with 43% now using AI to research or choose products alongside search engines and retailer sites, and 37% saying AI makes them more likely to try brands they had not previously considered. Then it found that only 7% of those shoppers think fast-delivery promises are consistently met.
That gap between 43 and 7 is the whole story. Agentic discovery is very good at introducing a shopper to a retailer they have never used before, which is exactly the moment when the delivery is the entire first impression. There is no prior relationship to fall back on, no years of "they usually get it right" goodwill. The first parcel is the brand. Locus is an AI-native logistics vendor and its conclusion, unsurprisingly, is that you need agentic fulfilment to fix this, so weigh the sales pitch accordingly. The 7% figure, though, does not need the pitch to be alarming. Locus's product chief, Pradyumna Chowdhary, put the consequence more plainly than the marketing usually allows.
This sounds clever right up until the agent actually places the order. Most retailer AI programmes that fail do so because the underlying data or operation was broken, not because the model was wrong, and this is the same lesson wearing a hi-vis jacket. You can win the discovery layer, get surfaced by every shopping agent going, and then torch a brand-new customer with a missed slot on their first order. We have seen the pattern before, in the dot-com years, when everyone built the shop window and worked out delivery later. The difference now is that the shop window is an agent that will happily send its owner somewhere else next time.
Where the money went
The venture market is already voting on where the hard problem is. UNIT AI raised $12m, co-led by Prologis Ventures, Dynamo Ventures and Ground Up Ventures, to scale a modular platform for end-to-end e-commerce fulfilment and returns. This is a US and North America story reading the coverage from over here, so no UK deployment is implied. What is worth clocking is the shape of the pitch: warehouse automation that drops into as little as 1,000 square feet, with a claimed return on investment inside twelve months. The footprint and the ROI are the company's own figures, so hold them lightly, but the positioning is the signal. This is money going into the boring half of the funnel while most of the attention stays on the shopping agent.
The returns angle is where the UK fashion read gets interesting. Returns are the tax on selling clothes online, and they are the part of the operation that agentic discovery makes worse, not better, because an agent that helps a shopper try three brands they had never considered is also an agent that helps them send two of them back. A small-footprint, modular returns-and-fulfilment robot is a genuinely more interesting proposition for an ASOS or a Boohoo than another shopping assistant. Nobody has shipped one to a British fashion warehouse yet. On this week's evidence, it is a better use of a returns director's attention than the fifth agentic-checkout demo of the quarter.
So the week ends honestly. It was a light one, three developments and no marquee UK launch, and it would be easy to file it as a slow news week. That would miss the shape. The industry spent the quarter making itself discoverable, and this was the week the cost of that showed up at the loading dock. Vercel says the agent is now the front door. Locus says the back door is jammed. UNIT AI says there is a business in fixing it. Being found by an agent was never going to be the hard part. Keeping the promise the agent made on your behalf is.
What to do this week
Sources
Discovery already moved to the agent
Vercel-commissioned research; May 2026 fieldwork, 4,000 European shoppers, 1,350 UK. Figures are the vendor's own. Distinct from our March-2026 Vercel London recap at /articles/vercel-agent-era-london.
The promise the warehouse can't keep
Locus research, 2,000 UK shoppers. Locus is an AI-native logistics vendor; the 7% delivery-promise figure and the 43%/37% adoption figures are survey findings, the 'agentic fulfilment is the answer' framing is the vendor's. The 85%-hesitate-after-one-poor-delivery figure quoted in the piece is separate research attributed to nshift.
Where the money went
US / North America focus; no UK deployment implied. The 1,000-sq-ft footprint and sub-12-month ROI are the company's own claims. Round co-led by Prologis Ventures, Dynamo Ventures and Ground Up Ventures.
Stay Connected
Follow LLCommerce on LinkedIn
Get the latest AI commerce insights, analysis, and industry news delivered to your feed.
Large Language CommerceAuthors
