Stop Asking When AI Will Arrive — It Already Does the Boring Work

Every few weeks someone publishes a new AGI timeline. Experts argue. Owners read the headlines and quietly conclude that AI “isn’t ready yet.” That conclusion is wrong, and it’s costing you hours every week.

Wrong question, wrong wait

AGI, artificial general intelligence, is the idea of a machine that can do anything a person can. Its arrival date is genuinely uncertain. But that question has nothing to do with whether AI can handle your purchase orders today.

The right question is simpler: what does it do now, for the jobs my team repeats every day?

The answer, for most small B2B operations, is quite a lot. Document handling, job tracking, customer comms. The tools exist. They work. The only thing missing is someone pointing them at the right problem.

Where the real work sits

Talk to any ops manager at a 20-50 person firm and you hear the same list. Emails arrive with PDFs attached. Someone opens the PDF, reads it, types the relevant fields into a spreadsheet or an ERP, then emails a confirmation. That loop runs dozens of times a day. Nobody enjoys it. Nobody has time to fix it. So it stays.

That loop is exactly what current AI handles well. Not because it’s intelligent in some grand sense, but because it’s consistent: the information arrives in a known place, it needs to end up in another known place, and the steps between are the same every time.

One build we did for a UK supply-chain operator extracts structured data from incoming documents automatically. Conservative estimate: 30 minutes saved per document. Volume was doubling month on month when we last checked. That’s not a prediction about future AI, thats a thing running in production right now.

The barrier isnt a £200k platform

Most owners assume automation means replacing their core system. It doesnt. The wins almost always come from removing the manual glue around the system you already use.

Freight teams running CargoWise are a good example. CargoWise stays. You dont rip it out. You wrap a small agent around the edges: it reads the email, pulls the figures from the PDF, and pushes them into the right fields. The ERP never changes. The manual step disappears.

The same pattern holds for any ops team using any half-decent system. Keep the core. Automate the edges.

Five questions to find your first win

Before you talk to anyone about tooling, answer these honestly:

• Which job does the team repeat every day? Not weekly, not occasionally. Daily, without fail.
• Where does the information arrive? Email inbox, PDF attachment, spreadsheet, supplier portal. Be specific.
• Where does it need to end up? A field in your ERP, a row in a sheet, a message to a customer.
• What’s the current manual time cost? Count it. Five minutes per document times 40 documents a day is over three hours. That number matters.
• What’s the smallest automation that can do one pass safely? Not the whole workflow. One step, done reliably, with a human check on the output until you trust it.

That last point is the one most people skip. They want the full transformation. The first win is almost never that. It’s one expensive repeat, removed.

What ‘safe’ looks like in practice

A sensible first build reads an incoming document, extracts the key fields, and writes them somewhere you can review before anything is committed. Something like this output structure is typical:

{
  "document_type": "purchase_order",
  "supplier": "Acme Freight Ltd",
  "po_number": "PO-20481",
  "line_items": [
    { "sku": "CTN-004", "qty": 120, "unit_price": 14.50 }
  ],
  "delivery_date": "2026-08-01",
  "confidence": 0.94 // flag anything below 0.85 for human review
}

You review the output. If the confidence score is low, a person checks it. After a few weeks of clean runs, you extend trust and let it push directly. That’s the pattern: start narrow, verify, then expand.

The honest summary

AGI may arrive in five years or fifteen. It doesn’t matter for your operations right now. What matters is that the tools available today can already handle the document your team opens every morning, the tracking update your customer chases every afternoon, and the confirmation email someone types from scratch every evening.

Dont buy transformation. Automate one expensive repeat. The first AI win is almost always the manual job your team already hates.

If you want to map that job and work out whether a small build makes sense, start with a conversation. No pitch, just a look at what you’re actually dealing with.