Insight: the endless possibilities of automation in enterprises

Most enterprises don't have an automation problem. They have a visibility problem. The work that's ripe for automation is so normal, so woven into someone's Tuesday, that nobody thinks of it as a process at all. It just happens — slowly, by hand, every week.

I've spent the last few years building agents that quietly absorb that work. What follows isn't a list of tools. It's the way I decide what to automate, in what order, and what to leave well alone. Mock text for now, but the structure is real.

Where the time actually goes

Before writing a single line, I follow the work for a week. Not the work people describe in a kickoff call — the work that actually shows up in their inbox, their spreadsheets, the seven tabs they keep open. The pattern is almost always the same: a handful of high-judgement decisions buried under a mountain of copy-paste.

That mountain is the opportunity. The judgement isn't. A good automation pulls the two apart — it does the mechanical 90% and hands the human the 10% that needs a person.

Automation is a spectrum, not a switch

The instinct is to ask “can this be fully automated?” — and when the answer is no, to drop it. That's the wrong question. Almost nothing in an enterprise is fully automatable, and almost everything is partly automatable. The real question is where on the spectrum a task belongs:

# do the mechanical 90%, escalate the 10% that needs a person
for case in inbox.unprocessed():
    summary = agent.read(case)
    if summary.confidence > 0.9:
        agent.resolve(case)          # auto-handled
    else:
        human.queue(case, draft=summary)   # needs judgement

That single threshold — a confidence cut-off, a person on the other side of it — is the difference between an agent people trust and one they switch off in a week. Start conservative. Earn the threshold up.

The patterns that survive contact with reality

Plenty of automations look great in a demo and quietly die in production. The ones that last tend to share a few traits:

1. They keep a human in the loop where it counts. Not as a bottleneck — as a reviewer who can say no in one keystroke.
2. They show their work.Every decision an agent makes is traceable back to the data it saw. Black boxes don't get adopted.
3. They fail safe. When unsure, they escalate instead of guessing. A quiet escalation beats a confident mistake every time.

What changes when an agent owns the workflow

The interesting shift isn't speed — it's what becomes possible once the boring part is free. When onboarding a new partner takes minutes instead of days, you onboard ten times as many. When a report writes itself overnight, you start asking questions you'd never have bothered to ask by hand. Automation doesn't just compress the work; it raises the ceiling on what a small team can take on.

That's the endless part. Every task you hand off frees attention for the next one — and the next one is usually more valuable than the last. The compounding is the whole point.

If any of this maps onto a process you're staring at right now, I'd genuinely like to hear about it. The best automations I've built started as someone describing the thing they dreaded doing on Monday.