The future I sketched in 2024 is no longer hypothetical. It’s shipping, it’s backfiring, and it’s rewriting who wins.

A follow-up to “Fashion Tech: The Present Future of Fashion and Technology” (October 2024)

When I published that first piece in October 2024, I argued that the blend of fashion and artificial intelligence was creating “limitless possibilities.” Roughly eighteen months on, the possibilities are no longer hypothetical. They are shipping. They are also forcing the harder questions the original post only gestured at: not just whether we can generate a model, a garment, or a storefront, but whether we should, and on whose terms.

The original throughline still holds. I compared today’s AI tooling to the sewing machine of the 19th century, a technology that reshaped how fashion is made, marketed, and consumed, mostly by collapsing time and cost. That frame has aged well. What I underestimated was the speed of the collapse, and how quickly the conversation would shift from efficiency to ethics, labor, and trust.

Here is where things actually landed.

The trajectory I called: hyper-personalization went from edge to baseline

In 2024 I described hyper-personalization, virtual fitting rooms, and digital “Sketch-to-Storefront” workflows as the leading edge. They are now closer to table stakes. By early 2026, McKinsey’s fashion technology outlook put adoption of machine learning for trend forecasting, planning, and 3D sample generation at roughly 48% of global brands. The AI-generated fashion photography market, barely a line item when I was writing, grew from about $1.51 billion in 2024 to around $2.01 billion in 2025.

The design tools I singled out, Browzwear and Lalaland.ai, did not just survive; they matured into a connected pipeline. Browzwear’s framing for 2026 is “idea to twin to shelf,” with the same digital twin now feeding not only fit and production but AI-generated marketing imagery. Virtual sampling, by various industry estimates, now cuts sample-development cost by 60 to 70% and time-to-market by up to half. The “first physical sample is the only sample” pitch I quoted in the original has, for a lot of teams, simply become true.

The proof point is no longer a vendor demo. MAS Holdings, the apparel manufacturer behind many of the intimates, swimwear, and performance brands you already know, (Victoria’s Secret, Nike, lululemon) began its digital product creation push in 2017 and stood up a dedicated Centre of Excellence in 2020.

It now develops more than 4,000 unique 3D styles a year for over 50 brands, with real-time co-creation replacing rounds of physical samples. The faster lead times and lower fabric waste arrived as byproducts of integration, not as one-off stunts. That is the version of “Sketch-to-Storefront” I was describing in 2024, finally operating at industrial scale.

Shift one: the synthetic model walked out of the back office and onto the cover

In the first piece I treated AI imagery mostly as a production convenience. The technology has since become good enough to be indistinguishable from a photograph, and that is exactly where the trouble started.

Mango ran AI-generated models in a 2024 campaign, with its CEO defending the move on speed grounds. Levi’s tested AI models and then walked the messaging back, insisting it was not a diversity strategy and that live shoots would continue.

Then, in August 2025, a Guess advertisement in Vogue, produced by the AI studio Seraphinne Vallora, became a genuine firestorm. The images were polished enough to read as editorial photography; only a small disclaimer revealed the model did not exist. The backlash was loud, and it was not really about image quality. It was about disclosure, displaced creative labor, and the narrow, conventionally “perfect” definition of beauty that synthetic models tend to default toward.

Regulation moved in parallel. The EU AI Act, adopted in 2024 and rolling out through 2027, pushes toward disclosure of synthetic media, transparency around training data, and respect for intellectual property, even when the underlying tools are open-source.

This is the part of the landscape closest to my own work, so I will be direct about the lesson. The defensible position is not “we used AI” versus “we didn’t.” It is whether you can show your work: where the training imagery came from, whether it was used with consent or under a clean license, and whether the audience was told.

Provenance stopped being a compliance footnote and became part of the product. The brands that weather the next backlash will be the ones that can answer “where did this image come from?” without flinching.

Shift two: the storefront itself is starting to dissolve

My 2024 piece assumed a shopper who visits a storefront, even a hyper-personalized one. The most disorienting development since then is that the visit may not happen at all. The shopper increasingly delegates to an agent.

The infrastructure arrived fast. OpenAI introduced an Agentic Commerce Protocol with Stripe; Google launched a Universal Commerce Protocol at NRF in January 2026; Microsoft shipped Copilot Checkout; Shopify rolled out agentic storefronts that let merchants sell inside ChatGPT, Copilot, and Gemini at once. Walmart and OpenAI announced a buy-in-chat partnership in October 2025, and Amazon extended its “Buy for me” capability while pointing its Rufus assistant toward comparison and autonomous purchasing. Shopify has said orders originating from AI-powered search grew roughly fifteen fold year over year.

And yet the reality check is just as instructive. OpenAI quietly paused its Instant Checkout, with fewer than 30 of Shopify’s millions of merchants live, conceding the experience lacked the flexibility it wanted. Walmart reportedly saw in-chat purchase conversion run about three times lower than sending shoppers to its own site. Forrester’s read on the moment was blunt: everyone has the fear of missing out, and nobody has actually figured it out yet.

For anyone building in this space, and I spend most of my week here, the takeaway is unglamorous but clear. The near-term winners are not the brands automating checkout most aggressively. They are the ones making their catalogs, inventory, pricing, and trust signals as readable by machines as by humans. The storefront is becoming an API. Fashion brands that still treat product data as marketing copy, rather than structured source-of-truth data, will simply be invisible when an agent comes shopping.

The map redrew itself, and not only because of technology

My original piece ranked the top online fashion retailers by 2023 sales: Shein at $14.4 billion, Walmart at $12.3 billion, Amazon at $8.4 billion. That snapshot is already a historical document, and the force that dated it was policy, not algorithms.

In 2025 the U.S. ended the “de minimis” exemption that had let sub-$800 parcels enter the country duty-free, the loophole that made the Shein and Temu model viable at scale. Prices rose, both companies warned customers directly, and transactional data showed price-sensitive shoppers migrating toward off-price department stores, secondhand apps, and domestic players. It is a useful corrective to any tech-first narrative: the consumer’s behavior is shaped at least as much by tariff schedules and trade policy as by virtual try-on.

The story so far is mostly digital, but one of the more interesting sustainability bets of 2026 is chemical.

In April, the Bezos Earth Fund committed $34 million to rethink what clothes are actually made of, on the logic that materials and manufacturing account for roughly 80% of fashion’s environmental footprint. The largest grant, $11.5 million, went to Columbia University in partnership with the Fashion Institute of Technology to grow a textile fiber from bacteria fed on agricultural waste: strong and breathable, compostable at end of life, requiring almost no land, and producing no microplastic pollution.

A fiber whose origin is a documented feedstock and a known biological process is provenance pushed all the way down to the molecule. The question the industry asks: “where did this come from?”, is now being asked of the thread itself.

The real divider isn’t AI adoption. It’s digital maturity.

Step back from any single tool and a pattern emerges. The brands getting caught flat-footed and the ones quietly compounding advantage are not separated by whether they use AI. They are separated by how deeply it is woven in.

A useful framing here comes from Browzwear’s digital-maturity work: maturity is not the act of adopting tools, it is integrating them into every part of the business until they drive continuous, scalable impact. Buying the software is the easy part. Rewiring how the organization works around it is the actual transformation.

That work sketches a five-phase climb, and it maps almost too neatly onto the stories above:

1. The analog trap. Traditional, siloed processes, no real strategy, digital as an afterthought.
2. Reactive experimentation. Isolated pilots run by individual teams, no executive sponsorship, wins that never scale past the team that ran them.
3. Intentional progress. A genuine digital strategy appears, often anchored by a dedicated Centre of Excellence, but adoption is still patchy across functions.
4. Integrated growth. Leadership-backed, cross-departmental, with digital embedded in day-to-day operations rather than bolted on.
5. Digital maturity. Agile, data-driven, customer-centric. Digital is no longer an add-on; it is the core the business runs on.

Seen through that ladder, a lot of last year’s headlines look less like innovation and more like Phase 2 wearing a Phase 5 costume. A splashy AI ad with a near-invisible disclaimer is an isolated pilot optimized for a press cycle, not an integrated practice with disclosure and provenance designed in from the start. The paused in-chat checkout was the same reflex at platform scale: ship the demo, skip the plumbing. Real maturity is comparatively boring. It looks like a manufacturer quietly producing thousands of digital styles a year because the whole organization, not one enthusiastic team, was rebuilt around it.

Provenance-clean, properly licensed training data is what lets you use AI imagery without inviting a backlash.

Here is where it connects back to my own throughline, and why I do not treat the ethics question and the agentic-commerce question as separate from this. The discipline that gets you up the maturity ladder, clean integrated data and a single source of truth, is the same discipline that makes you ready for both shifts.

Provenance-clean, properly licensed training data is what lets you use AI imagery without inviting a backlash. Structured, machine-readable product data is what lets an agent actually find and buy what you sell. Disclosure, provenance, and machine-readability are not three separate compliance chores. They are three faces of one mature operation that knows where its data comes from and where it goes.

Field notes from FIT Tech Week: the bleeding edge

I recently attended the Fashion Institute of Technology’s Tech Week event, listening to the founders actually building this layer, and the view from inside is more concrete, and more interesting, than the trend pieces suggest.

The sharpest demonstration came from Emily of Make the Dot, who walked through producing a denim collection called OSSA in roughly the time it usually takes to schedule a fitting. The old playbook she is replacing is familiar: watch what is selling, copy the winning trends, order in enormous quantities, and eat the waste. Her version inverts it.

An agent, “Dot,” aggregated the entire signal chain, from runway to brands to influencers to social to Google search trends, and generated design variations out of the pattern in that data. A human, Nicole, curated, deciding which pieces actually made the collection. Fit was tested digitally, and the agent produced a line sheet specified down to the wash, the whiskers, and the PP spray. Because the supply chain is vertically integrated, cotton to mill to cut-and-sew, a cycle she described as “six plus six” now runs in about four weeks.

Two of her lines stuck with me. “When making something small is nimble, a bet becomes a test,” she said, describing how a brand can float a social ad to gauge demand before committing a single yard to production. And the one that quietly reframes the whole sustainability conversation: “the path to the least waste also means more time for creativity.” She delivered it wearing the jeans.

That is the optimistic, operational story. The panel I sat in on, “Beyond the Headlines: The Real Ways AI Is Changing Fashion,” moderated by Rachel Sterling of The Pattern Maker and Alternew, pushed on where this goes next, and the founders mostly agreed the change is structural rather than cosmetic.

Franz Tschimben of ALLSIDES was candid that the specifics are hard to predict, but bet that 3D will emerge as a default mode of content creation as the cost of producing it collapses, the same digital-twin logic from earlier in this piece, generalized. Yusan Lin of Mirror Mirror AI described a decentralization of discovery: a world where anyone can be scouted, rather than waiting to be found by a gatekeeper. Sreya Halder of The Mall extended that to creation itself, a world where everyone gets to build their own village and curate it themselves.

Then Sophia Sterling, formerly of Google Creative Lab and now building a stealth venture called Paprika, planted the flag I found most worth carrying home. The company is named for the Diana Vreeland line, “a little bad taste is like a nice splash of paprika,” her argument against the tyranny of no taste at all. Sterling’s pitch is for what lives outside the algorithm: the human, physical, slightly-wrong instinct that recommendation engines flatten on contact.

It is the same tension the rest of this piece keeps circling. Dot can aggregate every signal in the market, but Nicole still decides what is good. The tools democratize who gets to make and be seen, and in the same breath they raise the value of the one thing they cannot generate, a point of view. If the last eighteen months were about proving the machinery works, the bet these founders are placing is that the next eighteen are about taste.

What this means for “the present future”

The sewing machine analogy I leaned on still works, but it needs a second half. The sewing machine made garments faster and cheaper, and in doing so it created entirely new questions about labor, standardization, and who got to call themselves a maker. AI is repeating that pattern at compressed speed across the whole pipeline, from concept render to synthetic model to autonomous checkout. The efficiency is real. So are the questions.

If the 2024 story was “limitless possibilities,” the 2026 story is that the possibilities now carry a price of admission, and that price is maturity: disclosed AI, consented and licensed training data, traceable provenance, and honest, machine-readable product information, all integrated rather than bolted on. The brands and builders who treat that as a constraint will keep getting caught flat-footed. The ones who treat it as the design spec are, I think, the ones who actually inherit the present future.

Joe Skopek is the founder of Velocity Ascent, an AI-first innovation consultancy based in New York and a member of the Leadership Team of the NYC Chapter of the NANDA Project from MIT Media Lab.

Sources and further reading

McKinsey & Company, The State of Fashion 2026: When the rules change (Nov 2025); McKinsey fashion technology outlook on ML adoption (early 2026)

Business of Fashion, on generative AI and virtual try-on (Jan 2026)

FASHN AI, Fashion AI: 7 Key Use Cases in 2026 (AI-generated photography market sizing, Feb 2026)

Browzwear, The Future of Digital Product Development: Trends Shaping Fashion in 2026 (“idea to twin to shelf”); Browzwear, A Digital Maturity Framework for Brands & Manufacturers (five-phase maturity model, MAS Holdings case study)

CNN, AI models in Vogue (Mango and Levi’s context, Jul 2025); Good Morning America / FashionNetwork, on the Guess-Vogue / Seraphinne Vallora campaign (Aug 2025)

Fast Company, Shop ’til you bot (agentic commerce, Instant Checkout pause, 2026); commercetools, The Agentic Commerce Radar (protocols, Walmart conversion data, 2026); Digital Commerce 360 (platform strategies, Apr 2026)

WWD and Morning Consult, on the end of de minimis and the Shein/Temu impact (2025); Business of Fashion, Will the End of De Minimis Kill the Shein Haul? (Jul 2025)

Bezos Earth Fund, Reinventing Clothes: $34 Million in Grants (Apr 2026); Columbia Engineering, on the $11.5M Columbia/FIT bacterial-fiber grant (Apr 2026)

FIT Tech Week (2026), author’s field notes: panel “Beyond the Headlines: The Real Ways AI Is Changing Fashion” (moderator Rachel Sterling, The Pattern Maker / Alternew; panelists Franz Tschimben, ALLSIDES; Yusan Lin, Mirror Mirror AI; Sreya Halder, The Mall; Sophia Sterling, Paprika), and the Make the Dot presentation (Emilie Ho, co-founder and CEO). Vreeland quote from D.V. (1984)

When your agents can’t prove they’re trustworthy, your business can’t either.

Everyone is racing to build smarter AI agents. Fewer people are asking the harder question: once we have trillions of them, how do they find each other, verify each other’s credentials, and collaborate without tearing the whole system apart?

Sandeep Mahendru – an Engineering Leader and Senior System Architect at Google, where he builds AI-driven solutions for the financial market – recently published a piece on Medium that lays out the clearest architecture I’ve seen for answering that question. It’s called Proposal for the Global Backbone of the Agentic Web, and it’s worth a careful read.

“We are currently building millions of specialized agents, yet we have no global address book to find them, no shared language to verify their skills, and no scalable security layer to ensure they aren’t actively sabotaging our workflows.”

I am walking through it in this post as it maps to work we’ve been doing at Velocity Ascent – and because the ideas in it have real consequences for anyone thinking seriously about where enterprise AI is heading.

The “Wild West” Moment

Sandeep opens with a frame I find hard to argue with:

“We are currently building millions of specialized agents, yet we have no global address book to find them, no shared language to verify their skills, and no scalable security layer to ensure they aren’t actively sabotaging our workflows.”

This is the gap. It’s not a technical gap in the narrow sense – the individual pieces exist. Large language models are remarkable. Agent frameworks are maturing. But the connective tissue between agents – the infrastructure that makes a network of agents trustworthy and useful rather than just chaotic – doesn’t yet exist at scale.

Sandeep calls this the “Cisco Moment” of the Agentic Web. The analogy is deliberate: the winners in the next era of AI won’t be defined by who builds the smartest model. They’ll be defined by who builds the backbone; the discovery, routing, and trust layer – that lets trillions of specialized agents find and work with each other safely.

* A “Cisco moment” is a stock market analogy referring to the dot-com era when networking giant Cisco Systems saw its stock plummet roughly 80% after telecom companies stopped buying its hardware. Today, financial analysts use the term to warn that massive capital expenditures in AI might create a similar bubble, particularly drawing parallels between Cisco in 2000 and Nvidia today. [1, 2, 3, 4]

From Code-First to Intent-First

The shift Sandeep describes isn’t just infrastructural – it’s epistemological.

Traditional software development is code-first. A developer translates a goal into explicit instructions, one procedure at a time. What’s emerging is intent-first orchestration: you describe what you want, and an underlying intelligent infrastructure figures out which combination of specialized agents can execute it.

“Instead of building monolithic software silos, developers are now focused on defining high-level goals that are executed by a ‘quilt’ of autonomous, specialized agents.”

The “quilt” metaphor is worth sitting with. Not a single God Model that does everything, but a federated patchwork of best-in-class experts – each sovereign, each specialized, each discoverable. The question is how you stitch them together in real time.

That stitching mechanism is what Sandeep calls the Brain.

The Brain: Semantic Intent Parsing at Scale

The Brain is the orchestration intelligence at the center of this architecture. Think of it less like a search engine and more like a very sophisticated dispatcher.

Traditional keyword search is too blunt for this. If a user’s request is “help me optimize our supply chain for Q4,” a keyword search returns documents. An intelligent Brain decomposes that intent into a structured, multi-agent workflow – logistics expert, finance agent, demand forecasting agent – and routes each sub-task to the best available specialist in real time.

Critically, Sandeep argues against a centralized bottleneck architecture:

“A standardized metadata routing protocol allows secure, local orchestrators to decompose human requests into multi-agent workflows while maintaining strict context isolation.”

This is a subtle but important design principle. Discovery doesn’t have to be centralized to be powerful. A federated model, where a lightweight metadata layer indexes agent capabilities across independently governed registries, can be both scalable and resilient.

The Discovery Gap: Why Agent Cards Aren’t Enough

The current standard for agent interoperability is A2A (Agent-to-Agent protocol), which gives every agent a structured “Agent Card” – essentially a business card declaring its capabilities, authentication requirements, and communication standards. A2A is built on established web infrastructure: JSON-RPC 2.0 over HTTP/S, OAuth 2.0, mTLS. It’s solid, enterprise-grade, and open.

But Agent Cards are self-declared. And self-declaration, as anyone who’s reviewed a vendor pitch deck knows, is not the same as verified performance.

AgentFacts: Report Cards, Not Business Cards

“Trust is no longer a ‘claim’ or a static audit; it’s an audited metric signed by third-party authorities and constantly tested via automated prompting and local consensus.”

AgentFacts are the proposed evolution beyond Agent Cards. Where an Agent Card says “I am a financial analysis agent,” an AgentFact is a cryptographically signed, W3C Verifiable Credential that says “this agent has been independently audited, has completed 12,000 successful tasks with a 97.3% accuracy rate, and has zero confirmed hallucinations in regulated output contexts.”

The distinction matters enormously in enterprise settings. When an agent is making decisions that touch compliance, financial transactions, or patient care, “trust me, I’m good at this” is not an acceptable credential.

Byzantine Fault Tolerance: In-Flight Security

Even cryptographically signed AgentFacts don’t solve the full security problem. Sandeep’s architecture adds a layer that I think most people building in this space haven’t adequately grappled with: Byzantine Fault Tolerance.

The analogy in the piece is the best summary I can offer:

“A Secure Federated Registry is like a ‘No Fly List’ – it keeps known bad actors out. But Byzantine Fault Tolerance is like ‘In-Flight Security’ – it protects the passengers even if someone with a valid ticket turns out to be a threat.”

BFT, in this context, means triangulating outputs across multiple independent agents and using quorum voting to detect when a “trusted” actor is hallucinating, has been compromised, or is simply wrong. Because traditional consensus mechanisms add latency, Sandeep proposes economic staking and slashing as the enforcement layer – agents put something at risk when they participate, and bad actors lose it.

This is crypto-native thinking applied to AI infrastructure. Whether or not you have a background in distributed systems, the principle is intuitive: skin in the game changes behavior.

Cognitive Liquidity: The AI-NPM Model

The long-term economic vision Sandeep lays out is what he calls Cognitive Liquidity – the ability to “import” verified intelligence as easily as a developer imports a software library today.

“By integrating discovered agents directly into low-code environments… developers can ‘import’ verified intelligence as easily as they import software libraries today.”

This is the AI-NPM model. Just as npm transformed software development by making reusable, versioned, community-maintained code packages trivially accessible, a mature Agentic Web would make reusable, verifiable, audited AI capabilities trivially accessible – streamable on demand, licensed transparently, and composable into complex workflows without custom integration work.

The infrastructure question isn’t just technical. It’s economic. Whoever builds the registry, the Brain, and the trust layer controls the tollbooth for an enormous amount of future AI commerce.

* In the context of AI, NPM (Node Package Manager) primarily refers to its role as the distribution hub for AI developer tools, libraries, and autonomous agents. While originally built for JavaScript, it has evolved into a “universal distribution platform” for the AI ecosystem.

Open Standards and the Public Good Mandate

One of the most important arguments in the piece is the one Sandeep makes almost as a footnote: this infrastructure needs to be governed by neutral bodies, not owned by any single platform.

A2A is already moving in this direction through the Linux Foundation and related open standard initiatives. NANDA – the Naming and Discovery Architecture – is one of the projects building the federated registry layer. If any single company owns the “Global Address Book,” the network effect becomes a chokehold. Open governance is the only model that scales without creating the kind of rent-extraction that would strangle the ecosystem before it matures.

A Note on Where This Work Lives

Sandeep Mahendru and I are both members of the NYC Chapter of the NANDA Project, an initiative out of MIT Media Lab focused on building the open naming and discovery layer for the Agentic Web. The work Sandeep describes isn’t purely theoretical – it’s the kind of architecture we’re actively thinking about in that community, and it’s influencing real infrastructure decisions being made right now.

The Agentic Web isn’t coming. It’s being built. The question is whether the backbone gets built right.

Why Agent Trust Matters to the C-Suite

If you’re a technology leader or founder thinking about AI strategy, Sandeep’s framework suggests a few reorienting questions:

• Are you building agents that will be discoverable by other agents? Or are you building another closed silo?
• How are you establishing verifiable trust in your AI outputs – not just internally, but in ways that external systems could validate?
• Are your AI investments positioned for the intent-first world, or optimized for the code-first world you’re leaving behind?

The companies that get this right early will have an enormous structural advantage. The backbone of the Agentic Web is being laid now, and the patterns being established will be hard to reverse.

Sandeep Mahendru’s original article, Proposal for the Global Backbone of the Agentic Web, is published on Medium. Sandeep is an Engineering Leader and Senior System Architect at Google, where he builds AI-driven innovative solutions for financial markets.

Joe Skopek is the founder of Velocity Ascent, an AI-first innovation consultancy based in New York. Both are part of the Leadership Team of the NYC Chapter of the NANDA Project from MIT Media Lab.