Last week, we spent three days at Web Summit in Lisbon — speaking with founders, operators and investors, and seeing hundreds of demos, workshops and main-stage sessions. Across it all, five themes kept resurfacing. Each theme points to a deeper shift in how retailers, brands and enterprise organisations will build, operate and compete over the next 18–36 months.
Below are the ones we believe matter most:
1. Agentic Commerce Moves from Theory to Deployment
Agentic commerce was everywhere this year — and not as a thought experiment. AI-driven traffic to ecommerce sites reportedly grew more than 4,700% year-on-year in July 2025, and retailers like Walmart and Etsy are already enabling transactions directly inside AI applications like ChatGPT and Perplexity. It’s a clear signal: a growing share of shopping journeys will soon begin with an agent, not a human.
That shift has consequences. If the primary “customer” becomes an AI agent optimising for structured data — price, stock accuracy, delivery terms, returns clarity — then the traditional discovery journey compresses. The battleground becomes data quality, not merchandising; clarity, not persuasion.
Dana Lawson’s session on Zero-UI engineering captured this well: retailers will need to design for AX (agent experience), not UX. That means cleaner schemas, machine-readable policies, real-time availability and well-structured APIs that let agents interpret and compare products confidently.
A wave of early infrastructure is emerging. EvoWeb lets brands create a dedicated AI-readable web layer. Ramonda AI provides a real-time MCP/API bridge so agents can fetch accurate availability, pricing and policy data — reducing the risk of hallucinated recommendations, out-of-date stock, or unsanctioned purchases.
But major questions remain. Agents still recommend out-of-stock or unsuitable products, largely because most aren’t yet integrated into live commerce systems. And as agents transact independently, intent, liability and dispute resolution become non-trivial. Retailers will need strong recourse models before this scales.
2. The Missing Layer in Enterprise AI: Context, Not Capability
Enterprise AI surfaced repeatedly, but the mood was clear: adoption is rising; ROI is not. Speakers cited MIT’s finding that roughly 95% of AI projects fail to scale, alongside McKinsey’s estimate that nearly 80% of companies using AI see no material bottom-line impact.
The message was the same across the board: the bottleneck isn’t model capability — it’s context.
Many enterprise deployments still sit on top of weak foundations: fragmented knowledge stores, shallow retrieval layers, inconsistent metadata and limited orchestration across systems. Without a shared, structured source of truth, even highly capable agents behave unpredictably, generating confident answers that collapse under scrutiny. In other words: enterprises don’t have a model problem, they have a memory and meaning problem.
A growing set of startups is positioning against this gap. KMapp, Contextual AI and KnowGuard AI each focus on strengthening the knowledge layer itself — building the contextual scaffolding that lets agents reason, retrieve and respond with reliability rather than surface-level fluency.
Until that layer matures, experimentation will continue to outpace impact.
3. Cybersecurity Becomes a Front-Line Commercial Risk
Cybersecurity felt unusually front-and-centre this year — especially for retailers, where the financial and reputational stakes are rising sharply. Recent breaches at Louis Vuitton, Dior and M&S illustrate how quickly incidents now escalate, with losses reaching tens or hundreds of millions. Cyber risk has now moved well beyond IT. It’s now a commercial issue.
The newest wave of startups reflects that reframing. Cyberly takes a human-centred approach, using AI-powered simulation and personalised training across email, SMS, voice, chat and deepfakes — acknowledging that most breaches originate with people, not perimeter failures. Whereas Hackurity covers the technical heavy lifting — automated pen-testing, round-the-clock monitoring and dark-web intelligence — aimed at organisations without a full internal security function.
A recurring theme: tools alone aren’t enough. Resilience — clear roles, practiced response plans and decision-making discipline — now matters as much as detection. In sectors built on trust, a breach doesn’t just compromise systems; it compromises brand credibility.
4. The Creator Economy Becomes a Performance Channel
The creator economy had a visible presence at Web Summit — not just in startup pitches, but across the headline speakers. Talks from Grace Beverley and Josh Richards underscored how far the space has moved: influence is no longer the product; it is the distribution engine for IP, brands and communities. Creators aren’t just producing content anymore — they’re building brands.
In parallel, we saw a new layer of tooling emerge to help brands run creator partnerships with more structure and less guesswork. WithGems is targeting Shopify-powered brands with automated creator matching, briefing and outreach. Trendin is pushing into enterprise-level campaign governance with workflow control, audience authenticity and analytics. MoonTech is taking a different angle, offering performance-based pricing where brands pay only for verified commercial outcomes.
What we’re seeing is a shift toward creators becoming a structured, performance-driven commercial channel. Brands that treat this as an ongoing, measurable capability — supported by the right platforms — are the ones unlocking repeatable value rather than sporadic campaign wins.
5. Robotics and Embodied AI Move Toward Practicality
Robotics and embodied AI were highly visible this year, from roaming floor demos to full humanoid models capable of shaking hands and holding conversations — including a striking blue-haired humanoid showcased by the Artificial Intelligence Association. But after the novelty wore off, the real question felt more practical than futuristic: what design, environment and ROI case will make robots useful, not just impressive?
Two of the strongest examples came from companies focused on operational efficiency rather than anthropomorphic design. Akara demonstrated how hospital-grade cleaning and workflow automation can unlock real clinical value, while Deliverance Enterprises showcased multi-purpose cleaning, delivery and building automation robotics for hospitality, logistics and smart infrastructure. Both prioritise safety, uptime, and integration over human-like interaction — and that felt far closer to deploy-ready reality.
At this stage, the robots gaining real traction are the ones built for a specific job and a clear business case — not humanoids designed to feel like co-workers. The bigger open question remains: as hardware capability accelerates, where will commercial value justify the investment, and what form factor will end-users truly trust — tool-like, human-like, or something in between?
From Hype to Hard Value
Web Summit gave a clear view of where the industry sits today. Curiosity has met capability, and the focus is shifting toward what creates repeatable operational value. Clean data, strong architecture, governance and clear ROI are becoming the real differentiators.
The takeaway wasn’t that everything is ready. It’s that the centre of gravity has moved. The next wave belongs to operators who can turn promising technology into measurable outcomes.
