The Agentic Commerce Manifesto:
Why E-Commerce is Dead

By Bill Chikirivao, Founder & CEO, Agentic Commerce

In the early 2000s, web-scale eCommerce wrecked the old database playbook. Relational SQL became deadweight; the rigid rows and columns simply couldn't keep up with the chaotic scale of Amazon's catalogs or the firehose of Google indexing. Out of that friction, NoSQL was born. Not Only SQL—a new way to structure data, forced by necessity.

Today, I think we're at a similar breaking point. We are entering the era of agentic commerce—where software agents don't just fetch options; they take the wheel. This is a fundamental shift where algorithms evaluate, negotiate, and execute purchases. Call it "agentic commerce," "autonomous shopping," or something we haven't named yet—but it's not a passing feature. Analysts are forecasting a multi-trillion-dollar economic shift over the coming decade.

For the past two years, the industry mandate has been simple: "just embed everything." We assumed vector search and RAG were the endgame. But we're learning the hard way that vectors are effectively the new flat files. They provide proximity—telling you what is nearby—but they lack the explicit structure required for complex reasoning.

We're approaching the "NoVEC" moment—Not Only Vectors. Like NoSQL, this doesn't mean the death of the old primitive. SQL is still ubiquitous, but it stopped being the only way to structure data. In the same way, vectors will remain a core building block for similarity search—but agentic commerce demands a richer semantic decision layer on top of them.

While Stripe builds the payment rails and frontier model providers build the foundational intelligence, neither is addressing the underlying data crisis. Anyone who has tried to make a serious agent choose between similar products has felt this wall. You simply cannot run a reasoning engine on a retrieval stack designed for static proximity.

We need a new topology.

The Cognitive Mismatch

At the root of the problem is a cognitive mismatch between how our current infrastructure is built and how agents actually reason.

When a human shops for running shoes, they perform a messy, intuitive calculus. They look at a photo and think, "Those look fast." They see a price and rationalize, "It's a stretch, but I'll wear them every day." They combine comfort, risk, social signaling, and budget into a fuzzy but real decision function.

An AI agent powered by frontier models—GPT-class systems, Claude, and the like—doesn't "browse" in this way. It reasons. It operates on vectors, constraints, and high-fidelity trade-offs.

Now imagine that agent has a rich, multi-dimensional intent, such as: "Find a shoe for humid marathon training, under $150, that balances durability with breathability and minimizes injury risk."

Today's commerce stack forces that intent to be squeezed into a crude keyword string or a single fuzzy embedding. Then we ask a vector index or search API—originally optimized for human-in-the-loop search—to stand in as the primary substrate for autonomous decision-making.

We're asking superhuman pattern recognizers to communicate via smoke signals.

Beyond Keywords and Vectors: The NoVEC Moment

To fix this, many teams are deploying vector search as the supposed savior. The assumption is simple: if we just embed the catalog, the AI will figure it out.

That's a half-measure.

Vector search is powerful, but it still maps a frozen territory. It finds similarities based on proximity in embedding space; it doesn't reshape the underlying distance between the user's mission and the product's reality.

In practice, that means:

• The agent can find "more like this,"
• But it struggles to reason about "best for this mission, under these constraints, with these trade-offs."

The NoVEC perspective is: vectors remain essential, but they are not sufficient. Agentic commerce needs a layer that treats vectors as one signal among many, inside a more explicit reasoning substrate—a topology we can inspect, tune, and explain.

That brings us to decision geometry.

The Fold: From Search to Decision Geometry

In the legacy web, products live on a mostly flat relevance plane. A query nudges you around that plane, and ranking functions decide which items you see first.

In agentic commerce, that is no longer enough. Products must exist within a multi-dimensional semantic topology shaped by missions, constraints, and outcomes. I use the metaphor of "the fold," borrowed from the string-folding scene in A Wrinkle in Time, to describe this.

In that metaphor, space is represented as a string: two points may be far apart along the string, but a "fold" brings them into direct contact. The path doesn't disappear—it just bends so that distance is redefined by a higher-dimensional geometry. In other words: the map bends so what matters most ends up closest to the agent.

That is what agents need.

In this new architecture, "search" is replaced by triangulation:

• The agent doesn't walk the catalog string looking for matches.
• The string folds so that the right region of the catalog comes to the agent's mission.

Concretely, the agent measures the semantic distance between the user's mission and each product's reality across multiple dynamic axes, such as:

• Signal Alignment – How closely does this product match the expressed and inferred mission?
• Contextual Price Position – Given this user and this mission, is the price a stretch, a bargain, or irrelevant?
• Mission Constraints – Does it violate hard rules like delivery window, region, allergies, or compliance?
• Outcome Risk – How likely is this choice to disappoint, given historical behavior, similar buyers, and downstream signals like returns or complaints?

This isn't just a "better filter." It's a shift from static retrieval—whether keywords or frozen vectors—to mission-adaptive reasoning over an explicit decision geometry.

Vectors remain critical as dense summaries of unstructured signals. In agentic commerce, they just have to sit inside a richer fold, not pretend to be the whole substrate.

The Cognitive Handoff

We are living through a cognitive handoff.

Until now, the buyer has shouldered most of the cognitive load: reading reviews, comparing options, checking specs, and rationalizing trade-offs. The software's job was to show options and gently nudge.

Agentic commerce inverts that.

In 2020, the stack was optimized to:

• Rank products in a grid
• Personalize just enough to boost conversion
• Let the human do the final reasoning

In 2026 and beyond, the expectation becomes:

• The agent does the heavy reasoning,
• The human approves or corrects.

A shopping agent might say:

"You told me you're training for your first marathon, you over-pronate, you live in a humid climate, and you strongly prefer to avoid returns. I've shortlisted two options and recommend Option B."

That requires a level of data fidelity most current merchant feeds simply cannot support.

Product feeds today are often unstructured chaos—SKU spreadsheets and marketing copy aimed at catching human attention, not enabling machine reasoning. In my own work, I keep seeing feeds where the questions humans care about most—will this break, will it arrive on time, will I regret this?—aren't modeled anywhere. They rarely express trade-offs, constraints, or outcome guarantees in a structured way.

For an agent's decision to be trusted, it needs defensible reasoning. Even a rough explanation could look like:

"I selected Option B because, while Option A was $20 cheaper, Option B scored 40% higher on durability and 25% higher on injury-risk reduction, which you specified as your primary constraints."

You cannot extract that explanation from a raw similarity score.

Vectors provide a number—a distance—but they don't inherently tell you why two things are close, or how that closeness relates to the user's mission. Only a semantic reasoning layer, grounded in explicit decision geometry, can provide the transparency required for trust.

The Invisible Economy

The implications for retailers and brands are stark.

In the next five years, a lot of homepages will start to feel like relics. A growing share of transactions will be headless and interface-less—initiated, negotiated, and completed by agents talking to APIs and decision services behind the scenes.

The winners in that world will not be the brands with the slickest UX. They will be the brands with the highest semantic resolution:

• Product data structured so that agents can "read" value propositions without friction
• Trade-offs, constraints, and outcomes expressed in machine-friendly ways
• Decision geometry exposed clearly enough that agents can confidently prioritize, justify, and learn

If your store is a black box of keywords and ad copy, you will be invisible to the agents. If your store exposes a clean decision geometry—machine-readable signals about suitability, risk, constraints, and value—you will win the algorithmic shelf space.

The rails and the agents are being built by the giants. Payments, identity, and foundational models are rapidly commoditizing.

The missing piece is the reasoning layer—the semantic topology that lets agents decide which product is right for which mission and explain why.

It's time to stop treating vectors as the end state, close the search box, and unfold the geometry.