LEAD-01-004-P · Primer — Why Now

The idea in one sentence

Agents became viable this decade because four capabilities matured at once — tool use, long context, cheap reasoning, and reliable function calling — turning a clever text generator into something that can actually get work done.

Why it exists

Leaders rightly ask why agents did not work five years ago and whether this is just a temporary hype cycle; the honest answer is that specific technical thresholds were crossed only recently. Understanding what changed protects you from both dismissing agents as vaporware and over-trusting them as finished. It is like asking why ride-hailing appeared around 2010 and not 1995 — the idea was obvious, but GPS, smartphones, mobile payments, and cheap data all had to arrive together before it could exist.

How it works

1. Tool use: models can now call external systems — databases, APIs, calculators, browsers — so they act on the world instead of only describing it. 2) Long context: a model can hold an entire contract, data room, or codebase in working memory, so it reasons over real business material rather than fragments. 3) Cheap reasoning: the cost per unit of thinking has fallen by orders of magnitude, so an agent can afford to plan, try, check, and retry many times per task. 4) Reliable function calling: models emit structured, machine-readable requests dependably enough to wire into production systems. Remove any one and the agent collapses — no tools means no action, no context means no memory, expensive reasoning means it is uneconomical, and unreliable function calling means it cannot be trusted in a pipeline. A procurement agent needs all four: reading a long vendor contract (context), computing spend against budget (reasoning), pulling live pricing from an ERP (function calling), and submitting the requisition (tool use). These four converging is exactly why the delegation shift is happening now and not a decade ago.

See it

flowchart TD
  A["Tool use<br/>acts on world"] --> E["Viable<br/>agent"]
  B["Long context<br/>real memory"] --> E
  C["Cheap reasoning<br/>plan and retry"] --> E
  D["Function calling<br/>wires to systems"] --> E

In practice

A finance-close agent is only possible because it can hold the full ledger and prior-period workpapers in context, reason cheaply enough to test dozens of reconciliation hypotheses, call the ERP through reliable function calls to fetch balances, and use tools to post adjusting entries into the accounting system.

The one thing to remember

Agents work now — not because the idea is new, but because tool use, long context, cheap reasoning, and function calling finally arrived together.