The end of the billable hour

This essay is about our own industry, so let us be honest about what that industry was. For decades, IT services built empires on a single arbitrage: hire engineering talent where it costs less, bill it where it costs more, and meter the difference by the hour. Time and Materials, staff augmentation, body shopping — the names varied, but the unit of sale never did. The unit was the hour. And the hour was a fine unit for as long as every technology that entered the industry still consumed hours: new languages, new frameworks, new clouds — each wave changed what the hours were spent on, never how many were needed. This wave is different in exactly one way that turns out to be the only way that matters.

Call it the GenAI paradox. If a junior engineer working with AI agents can now do in two hours what used to take twenty senior hours, a firm that bills by the hour faces an impossible choice. Pass the efficiency to the client, and your top line shrinks by the same multiple your engineers improved. Keep the efficiency, and you are undercut within a quarter by leaner, AI-native shops with no legacy revenue to protect. Under T&M, inefficiency was quietly incentivised — more hours meant more revenue, and nobody had to say it out loud. AI did not merely expose that misalignment. It made it unaffordable, for both sides at once.

The billable hour survived every technology of the last fifty years, because every technology still needed hours. This one consumes them.

A model that priced the wrong thing

The deeper problem was never the rate card; it was what the rate card measured. T&M prices inputs — hours worked, full-time-equivalents deployed — when the client only ever wanted outputs: the feature shipped, the ticket resolved, the process automated. For decades the gap between those two things was where the industry's margin lived. Now that AI has commoditised the execution layer — the coding, the routine testing, the legacy maintenance that filled the timesheets — the gap is collapsing, and the pricing model is collapsing with it. What is being reinvented, urgently and everywhere, is the unit of sale itself: away from inputs, toward outputs and outcomes.

The three replacements

The first is outcome-based pricing: charging not for the build but for the profit-and-loss impact the build creates. Per cloud-cost dollar saved. Per support ticket resolved autonomously. A share of the revenue lift from an algorithmic pricing engine. It requires discipline that T&M never demanded — mutually agreed success criteria nailed down before a line of code is written, and measurement infrastructure both sides trust — because the vendor is no longer a supplier of effort but a partner in risk.

The second — and the one we believe wins — is services as software. Instead of building everything bespoke from scratch, from zero, every time, the firm builds a modular, proprietary platform: the agentic core, the memory, the integration fabric. Clients subscribe to the platform and pay a far smaller professional-services fee for the customisation that makes it theirs. The mindset shift is from "build for me" to "subscribe to mine": the firm finally earns software margins on a services relationship, and the client gets in months what bespoke-from-zero delivered in years.

The third is the virtual FTE: instead of renting a human developer, the client pays a flat monthly fee for a blended unit of AI agents plus human oversight, with a guaranteed volume of output — a tier that includes, say, fifty minor feature shipments or compliance checks a month. It keeps the familiar rhythm of a staffing contract while quietly replacing its substance: what is being rented is capability, not people.

The giants are already turning

This is not a boutique theory; the largest ships in the industry are turning. Infosys now generates roughly 54 percent of its revenue from fixed-price contracts, and Cognizant has scaled its own fixed-price share to about 47 percent — structures in which AI velocity fattens the margin instead of shrinking the invoice. The same giants are building proprietary platforms — Infosys Topaz, TCS WisdomNext — cognitive layers they place on top of client legacy systems for recurring, software-like revenue. At the top of the pyramid they are carving out elite advisory units that counsel boards on AI restructuring at premiums that offset the dying maintenance layer. And in service desks, the rewrite is already literal: contracts that once billed per helpdesk seat now deploy agentic AI to auto-resolve tier-one issues and bill exclusively on successful resolutions within SLA.

When the largest labour brokers on earth stop selling labour, believe them.

Why the Titanic turns slowly

If the destination is so obvious, why is the voyage so hard? Because everything inside a legacy services firm was built for the old unit. Revenue first: T&M produced predictable, recurring monthly cash flow, and outcome-based contracts replace it with peaks and valleys that public markets punish. Attribution second: if a feature is priced on sales lift and the client's marketing team runs a terrible campaign, whose miss is it? Disentangling that requires measurement infrastructure most engagements have never had. Culture last, and hardest: delivery organisations are still managed on utilisation and timesheets — and a timesheet-driven management layer cannot bring itself to reward an engineer for automating away a billable hour. That is not a pricing change; it is a corporate behavioural rewrite.

Which is why the transition favours firms with less to unlearn. A boutique with no installed base of hours to protect can price the new way from the first contract — and this is the honest reason AI-native shops are winning work from firms fifty times their size.

Where we placed our bet

KRDS is a services company, so none of this is commentary from a safe distance — it is the water we swim in, and services as software is the bet we have already placed. Our platform is the proprietary core: the agentic layer, the institutional memory, the connective tissue between a client's systems. Clients subscribe to the system; the bespoke work — the part that makes it fit their organisation and no one else's — is the smaller, sharper engagement on top. And because the system learns from its own operation, it needs fewer hours every month it runs, which under the old model would have been a problem and under this one is precisely the product. The firms that survive the next few years will stop acting like labour brokers and start acting like business-risk partners — sharing the downside when things fail, capturing real upside when AI-accelerated systems deliver outsized returns. You cannot bill by the hour for a system whose entire purpose is to need fewer of them. So we stopped selling hours. We sell a system that stays, and learns — and gets cheaper to run precisely because it does.