Your Agency Still Bills by the Hour. Computing Doesn't Work That Way Anymore.

A three-person engineering team at StrongDM targets $1,000 per day in AI token spend. They write no code by hand. They do no traditional code review. Their output exceeds what the team produced before. The CTO, Justin McCarthy, disclosed this publicly in early February 2026, and the reaction from most of the technology press was predictable: breathless excitement about the future of AI-assisted development.

The reaction from the ecommerce agency world was silence. It should have been panic. Because that $1,000 a day isn't just a budget line — it's a unit of measurement. And when the unit of measurement for computing shifts from hours of human labour to dollars of purchased intelligence, every business model built on billing human hours faces an existential problem that no amount of "AI-enhanced services" marketing can fix.

The Unit of Work Changed. You Probably Missed It.

For sixty years, the fundamental unit of computing was the instruction. A human wrote code, a machine executed code, and value was denominated in how cleverly the human could sequence those instructions. The developer's job was translation: convert business logic into machine logic, one function at a time, one ticket at a time. The agency model was built on top of this reality. Clients paid for human time because human time was the bottleneck. An hour of a senior developer's attention was worth £150 because that hour produced instructions that machines couldn't produce themselves.

That unit of work has changed. The fundamental material of computing in 2026 is the token — a unit of purchased intelligence. You don't tell the machine what to do step by step. You describe what you want, feed it context, and buy enough intelligence to get a result. The machine figures out the workflow on its own. OpenAI is reportedly planning to charge $2,000 to $20,000 per month for AI "employees" — agentic systems that can do research, write software, and execute complex workflows autonomously. Those aren't tools. They're substitutes for the very labour agencies sell.

This isn't a tools upgrade. It's not even a productivity enhancement. It's a change in what computing is. And once you see it, you can't unsee the implications for any business model that derives revenue from selling human hours of technical work.

The Price Curve That Destroys Your Rate Card

GPT-4 equivalent performance cost $20 per million tokens in late 2022. Today it costs roughly $0.40. That's a 98% price decline in three years. And here's the part that matters for agencies: when a resource gets dramatically cheaper, people don't use less of it — they use enormously more. This is Jevons' paradox, a well-established observation that efficiency gains in resource consumption lead to increased total consumption, not decreased.

Steam engines got more efficient; coal consumption exploded. Cloud computing got cheaper; AWS bills went up. AI inference gets cheaper; the volume of intelligence being purchased is skyrocketing. The average organisation now spends $85,000 per month on AI — up 36% year over year — and the share planning to spend over $100,000 monthly has doubled from 20% to 45%.

Now apply Jevons' paradox to the agency market specifically. As the cost of building software collapses, the total volume of software being built will explode. That sounds like good news — more demand! Except the demand isn't for agencies. It's for tokens. A client who previously needed a six-person agency team for a Shopify migration can now accomplish the same work with one competent person managing AI agents and a $3,000 monthly token budget. The total spend on the project might be similar, but the distribution of that spend has shifted from human labour to machine intelligence. The agency's rate card, denominated in hourly human time, has become the wrong unit of measurement for the value being created.

Three Developer Tracks — Three Threats to Your Agency

The developer world is differentiating into three distinct tracks, and each one represents a specific competitive threat to the traditional ecommerce agency.

The Orchestrator is a developer who writes no code but specifies outcomes and manages the intelligence that produces them. Their skills are system design, specification writing, quality evaluation, and token economics. They think in agent architectures, context windows, evaluation frameworks, and cost per outcome. One orchestrator managing a fleet of AI agents can produce the output of an entire agency development team — at a fraction of the cost and a multiple of the speed.

For agencies, this is the most immediate threat. The orchestrator doesn't need your team. They need a token budget. A single orchestrator with $1,000 per day in compute can outproduce a ten-person agency team billing collective hours. The maths is brutal: your senior developer at £150/hour produces maybe 6 billable hours of output per day, or £900 in billed value. The orchestrator's $1,000 in tokens produces output that previously required multiple developers working for weeks.

The Systems Builder constructs the infrastructure that orchestrators use — agent frameworks, evaluation pipelines, context management systems, model routing layers. This is deep technical work with a high compensation ceiling. For agencies, systems builders represent a talent drain: your best engineers are leaving to build AI infrastructure because the upside is higher and the compensation reflects it. The agency talent pipeline is being siphoned at the top end.

The Domain Translator is the threat nobody sees coming, and it may be the most destructive of all. These are professionals who combine enough technical fluency to work with AI systems and enough deep domain expertise to know which problems are worth solving. The dental practice management specialist who can now build software tools. The construction scheduling expert who can now automate workflows. The ecommerce operations manager who spent fifteen years learning inventory management and fulfilment logistics — and can now build the tools she previously hired agencies to configure.

This is the person who knows your client's business better than you do, and now has access to the same building capability you sell. She doesn't need to hire a developer. She doesn't need to hire an agency. She needs a £200-per-month Claude subscription and a weekend.

The Domain Expert Ate Your Lunch. You Didn't Notice.

The domain translator threat deserves special attention because it represents a category of competition that agencies have never faced before.

Traditional competitive dynamics in the agency space were straightforward: agencies competed with other agencies. The barriers to entry were capital (hiring developers is expensive), expertise (building ecommerce systems requires specialised knowledge), and relationships (clients trust established firms). All three barriers are collapsing simultaneously.

Capital: the cost of building has plummeted. A meaningful ecommerce integration that would have cost £50,000 in agency fees two years ago can now be built with £2,000 in token costs and a competent orchestrator. Expertise: AI systems carry the technical knowledge that previously resided exclusively in senior developers' heads. Relationships: the client's own operations team, armed with AI tools, has a deeper relationship with the business than any external agency ever could.

Consider a concrete example. A mid-market Shopify merchant with 50 locations needs a custom inventory management integration. In the old model, they'd engage an agency at £120-£180 per hour, scope a six-week project, and pay £40,000-£60,000 for a team of developers to build it. In the token economy, their operations director — who understands the inventory logic better than any external developer — sits down with an AI coding assistant, describes what she needs, iterates on the output, and ships a working integration in a week. Total cost: perhaps £500 in AI subscription fees and her own time.

The quality might be 80% of what the agency would deliver. But 80% quality at 1% of the cost, delivered in one-sixth the time, by someone who actually understands the business requirements first-hand? That's not a close call. That's a rout.

And it's precisely the dynamic that's playing out across the Shopify ecosystem right now. Merchants who previously hired agencies for every customisation are discovering that AI tools let them solve their own problems — imperfectly, but fast and cheap. The agency doesn't even get the opportunity to compete because the client never issues an RFP.

Jevons' Paradox Hits the Agency Market

Here's where the second-order effects get properly uncomfortable. Cheaper intelligence doesn't just help your existing competitors. It creates entirely new ones.

When building software required a team of experienced developers, the number of agencies capable of delivering complex ecommerce work was naturally limited. Entry barriers were high. You needed talent, which was scarce and expensive. You needed processes, which took years to develop. You needed a portfolio, which required surviving long enough to accumulate one.

Token-based computing demolishes all three barriers. Talent? One competent orchestrator can produce agency-grade output. Processes? AI agents follow specifications with machine precision — no years of institutional learning required. Portfolio? When building is cheap and fast, new agencies can produce impressive demo work in days, not years.

The result is that the number of entities capable of delivering "agency-grade" ecommerce development is about to multiply dramatically. Not by 2x or 5x, but by orders of magnitude. Every freelancer with domain knowledge becomes a one-person agency. Every operations manager becomes a potential competitor. Every AI-savvy marketer becomes capable of building integrations that previously required a development team.

A16Z's data shows that AI-native companies operate at 3–5x the revenue per employee of traditional SaaS companies. A $10 million ARR startup that might employ only 15 people, compared to 55-70 at a traditional equivalent. Apply that ratio to agencies: the AI-native agency that produces the same output as a 30-person traditional agency, but with 6-8 people and dramatically lower overhead, can undercut on price while maintaining margins. Every market niche that supported three agencies will soon support thirty. And most of those thirty won't look like agencies at all — they'll be consultants, solopreneurs, and domain experts who happen to have AI tools.

Token Economics Is Now a Core Business Competency — and Most Agencies Don't Have It

Cursor, the AI coding editor, became a billion-dollar revenue company extraordinarily fast — and then nearly imploded when Anthropic introduced priority service tiers and raised prices. Cursor was sending essentially all of its revenue to Anthropic in API costs. When those costs spiked overnight, Cursor was forced to gut its unlimited $20-per-month plan and introduce a $200-per-month tier. Users revolted.

The lesson isn't that tokens are expensive. The lesson is that token economics is now a core business competency, and companies that don't master it are one supplier pricing change away from crisis. Cursor's eventual response was to build its own model — an acknowledgment that depending entirely on a third-party intelligence supplier is an existential business risk.

Agencies face an analogous challenge. If your business model depends on AI tools you don't control, your margin is at the mercy of whoever sets the token price. But the deeper problem isn't supplier risk. It's that most agencies have no framework for thinking about intelligence as a variable cost that can be optimised, routed, and managed. They treat AI as a tool their developers use, not as a resource that needs its own operational discipline.

The agencies that will survive the token economy are the ones that develop genuine competency in intelligence operations: routing different tasks to different models at different price points, building evaluation frameworks that measure output quality per token spent, creating specifications precise enough that AI systems can execute them reliably, and constructing quality gates that catch errors before they reach clients.

This is a fundamentally different skill set from traditional agency management. It's closer to manufacturing operations than to creative services. And most agencies are culturally, structurally, and financially unprepared for the transition.

What Token-First Agencies Actually Look Like

The agencies that restructure successfully around the token economy will be nearly unrecognisable compared to their predecessors.

Pricing restructures around outcomes, not inputs. Instead of billing £150 per hour, token-first agencies price based on the value of the deliverable. A Shopify migration costs £X. A custom integration costs £Y. The agency's internal costs — whether human or token — are invisible to the client. This requires agencies to develop much sharper competency in scoping, estimation, and value pricing, skills that most agencies have historically avoided because hourly billing was easier.

Team structures compress dramatically. The 30-person agency becomes a 10-person agency with a larger token budget. Senior roles shift from "best coder" to "best specifier" — the person who can write the clearest brief for an AI system, evaluate the output most effectively, and manage the intelligence budget most efficiently. Junior developer roles don't disappear, but they transform into quality assurance and specification refinement positions.

Domain specialisation becomes the primary differentiator. When building capability is abundant, knowing what to build becomes the scarce resource. The agency that deeply understands ecommerce fulfilment logistics, or subscription commerce economics, or B2B marketplace dynamics, has a durable advantage that no amount of token spend can replicate. Generic "full-service digital agencies" are the most exposed category because their value proposition — we can build anything — is exactly the claim that AI tools now make for free.

Client relationships restructure around continuous intelligence delivery. Rather than project-based engagements with defined start and end dates, token-first agencies offer ongoing operational intelligence: monitoring, optimising, and evolving their clients' ecommerce systems continuously. The engagement looks less like a construction project and more like a managed service — which, not coincidentally, is a model with better margins and more predictable revenue.

The Question That Actually Matters

The Silicon Valley discourse around the token economy is obsessed with scale: $20,000-per-month AI employees, billion-dollar solo-founder companies, $285 billion in infrastructure spending. That's the venture capital perspective. It's not the one that matters to the ecommerce agency owner in Manchester or Melbourne trying to figure out whether their business model survives the next eighteen months.

For that person, the question is narrower and more urgent: when the fundamental unit of computing changes, what happens to a business model built on selling the old unit?

The honest answer is that it breaks. Not immediately, not uniformly, but inevitably. Hourly billing for technical services was rational when human hours were the bottleneck. They're not the bottleneck anymore. Intelligence is now purchasable by the token, and the cost is falling at 98% per three-year cycle. Every hour you bill is competing against a machine that does the same work for tokens — and the token price is still dropping.

The agencies that thrive in the token economy will be the ones that stop selling hours and start selling outcomes. The ones that develop genuine expertise in intelligence operations rather than treating AI as a productivity feature. The ones that specialise deeply enough in specific domains that their knowledge — not their build capacity — is what clients pay for. The ones that restructure their teams around specification quality and output evaluation rather than code production volume.

The unit of work changed. Your rate card didn't. Fix that before your clients do it for you — by firing you and buying tokens instead.