The 3-Pronged Fork
Let's start from the basics.
Why do we need computers at all?
To make life easier. To make things faster. To make things more accurate. To help us make decisions faster than we could on our own.
Those same reasons have only gotten sharper and more competitive as the technology industry has evolved. Speed and accuracy stopped being nice-to-haves and became the edge - the difference between a company that wins a market and one that gets quietly outrun. Digitization was such a welcome shift precisely because it shrank the whole world down into our palms.
And with that came the appetite to automate more, and more, and more. We wanted everything to happen at the snap of a finger.
It's worth remembering: whatever feels effortless to the user is almost always the product of a fairly complex design sitting quietly in the background, invisible to the person using it. In 1999, Bill Gates wrote an entire book about this shift toward running a company at digital speed - Business @ the Speed of Thought - arguing that the companies that would win were the ones that could move information, and decisions, as fast as they could think them. That was a fairly visionary thing to say in 1999, and it's essentially the promise agentic AI is now trying to cash a quarter-century later.
Which is also why the gap between the pitch and the deployment is worth naming honestly. Gartner expects task-specific AI agents to show up in 40% of enterprise applications by 2026, up from under 5% in 2025 - a genuinely fast climb. But McKinsey's 2025 State of AI survey found that in any given business function, no more than 10% of organizations are scaling agents in production, even though 23% have started experimenting somewhere. Plenty of pilots. Not nearly as much of it sticking.
That gap is usually a symptom of the same underlying question, unanswered: what the business world really wants out of all this is more money. Fair enough - automate or don't, shareholders want their piece of the pie. But set that aside for a moment and ask a more useful question: who does automation truly impact, especially inside an enterprise? Is it the customer? The management layer? The workforce doing the work?
The fork test
Here's what I call the 3-pronged fork approach to evaluating that question. It works as a framework for any enterprise automation or digitization initiative - arguably it belongs at the center of the business strategy itself, not bolted onto it afterward.
Most usefully, it's a clean filter for evaluating any agentic AI use case: does it help management decide or operate better, does it help the customer, or does it protect the team from burnout? If a proposed use case doesn't clearly serve at least one of the three, it's probably automation for its own sake rather than something worth building.
Prong one: Management - the value is leverage
Agents synthesize data across systems in real time, absorb operational volume without a linear increase in headcount, enforce policy the same way every time, and free leadership from firefighting toward actual strategy.
Moderna is a useful, well-documented example of what this looks like at scale. After rolling out ChatGPT Enterprise, employees built roughly 750 custom GPTs across the company within two months, with legal reaching full adoption and using a “Contract Companion” agent to handle first-pass contract review and summarization. Moderna's CEO, Stéphane Bancel, summed up the intent bluntly: “If we had to do it the old biopharma ways, we might need a hundred thousand people. We really believe we can maximize our impact with a few thousand using AI” — leverage, not just speed (OpenAI).
Prong two: Customer - the value is availability and continuity
No queue time, no time-zone gap, full context carried into every interaction so service feels personal rather than scripted, and agents that can flag a problem before the customer ever files a ticket.
Klarna's AI assistant is the case everyone cites here, for good reason. Within its first month it handled 2.3 million conversations - two-thirds of the company's customer service chats - across 23 markets and more than 35 languages, around the clock. Klarna said that was the equivalent work of roughly 700 full-time agents, and average resolution time dropped from 11 minutes to under 2 (Klarna; OpenAI). No queue, no time-zone gap, consistent answers at 3 a.m. as at 3 p.m. - that's the customer prong working as designed.
Prong three: Team - the value is subtraction, not addition
Agentic AI's best use on this prong is removing the repetitive, low-judgment work - status chasing, data entry, report compilation, overnight monitoring - so people get their attention back for the work that actually needs a human. This is the prong most enterprises skip. It's also the one this framework insists on: burnout reduction should be a designed outcome, not an accident.
JPMorgan's COIN platform is an early but instructive example. It was built to take over the manual review of commercial loan agreements - work that had previously consumed an estimated 360,000 hours of lawyers' and loan officers' time every year - and turn it into a task the system completes in seconds, freeing that team for judgment calls the software can't make (Bloomberg). More recently, Microsoft's 2026 Work Trend Index found that 66% of AI users report spending more time on high-value work as a direct result, rising to 80% among its most advanced “Frontier” users (Microsoft WorkLab) - which is exactly the subtraction this prong is meant to produce.
Where it gets interesting: the trade-offs
These three prongs can trade off against each other. Aggressive cost-cutting on the management prong, for instance, can quietly push more edge-case handling onto the team, undoing the very burnout benefit the initiative was supposed to deliver. A good agentic AI initiative should be net-positive across all three prongs - not optimized on one at the expense of another.
Klarna, again, is the clearest public example of this dynamic playing out in real time. The same AI assistant that looked like a management and customer-service win in 2024 - cutting headcount by roughly 22% during a year-plus hiring freeze while the bot handled about 75% of chats - became a cautionary tale by 2025, when Klarna started recruiting human agents again. CEO Sebastian Siemiatkowski told Bloomberg the AI produced “lower quality” support than people did, and that “investing in the quality of human support is the way of the future for us” (Entrepreneur). The management prong looked optimized - lower cost-to-serve, fewer heads. The customer prong quietly slipped. That's the fork test doing exactly what it's for: catching a use case that won on one prong while losing on another.
The takeaway
Run any proposed agentic AI initiative through the fork before you build it. Does it help management decide or operate better? Does it make things more available or more personal for the customer? Does it genuinely subtract repetitive work from the team, rather than just adding a new tool for them to babysit? If it clearly lands on one prong, that's a start. If it can't defend itself against the other two, it isn't a strategy yet - it's automation for its own sake, dressed up in agentic language.
Sources
Comments