Writing
The AI Automation Landscape: What Each Category Is Actually For
Five categories, honestly described, including the one we are in and what it is bad at.
Published · Updated · 9 min read
The short version
Connectors move data on rules. Agent platforms build reasoning systems you maintain. Assistants make individuals faster. Agencies build what you specify. Consultancies decide what to build and hand it over. Pick by which of those you are missing.
Every tool in this market is good at something, and the wasted money comes from buying a category that solves a problem you do not have.
Category 1: connectors
Zapier, Make, n8n, Pabbly Connect. They move data between applications when a trigger fires.
Genuinely excellent at deterministic plumbing, and cheap. Their limit is that they need the rule stated in advance, so any work involving judgement still stops at a human.
Buy if: you know what to automate, the logic is stated, and someone in-house will build it.
Category 2: agent platforms
Relevance AI, Stack AI, Gumloop and the rest. Model-first builders for systems that reason over your data rather than follow a rule.
The capability is real and the cost is low relative to services. The limit is that they hand you a workshop, not a finished thing, and the hard parts sit outside the tool: choosing the right target, preparing your history, and getting the team to trust it.
Buy if: someone technical will own agent design as an ongoing job.
Category 3: assistants
Lindy, Bardeen, and every copilot bundled into software you already pay for. They make one person faster at their own work.
Cheap, per seat, no project required. The limit is arithmetic: making eight people twenty percent faster does not remove an operation, and the improvement never appears on the P&L.
Buy if: individuals are drowning and you want relief this week.
Category 4: agencies
They build what you specify, usually on a retainer. Fast and cheaper than a consultancy on a well-defined scope, and some of them are very good.
Two limits. They take your diagnosis as the brief, and you are usually right about the symptom and wrong about the cause. And a retainer pays for continued involvement, which quietly shapes what gets built towards things that need attention.
Buy if: you already know exactly what you want built.
Category 5: consultancies that also build
This is us, so treat the description accordingly. Assess every operation, find the one costing the most, replace it, train your people, leave.
What it is bad at: it is slower to start than buying a tool, more expensive than an afternoon in Zapier, and pointless if you already know what to build and have someone to build it. If your bottleneck is capability rather than diagnosis, buy a platform.
Buy if: nobody in-house will own it, and you want the target chosen by cost rather than by curiosity.
Choosing by what you are missing
| What you are missing | What to buy |
|---|---|
| A rule between two apps | A connector |
| Time in one person's day | An assistant |
| A place to build reasoning systems | An agent platform |
| Hands to build a defined spec | An agency |
| A decision about what is worth building | A consultancy |
| Permanent capability in your product | A hire |
The mistake that costs the most
Buying capability when the missing thing is a decision. A platform subscription, a builder, and six weeks later a clever system attached to a cost line that was never the problem.
The reverse mistake is cheaper but real: hiring a consultancy for work a $40 tool covers. That is why an honest assessment sometimes ends with us telling a company to buy Zapier.
Questions
Can one tool cover all of this?
The categories are converging at the feature level, but the question is not which product has the widest feature list. It is whether the thing you are missing is a tool, a builder, or a decision.
What do most $5M companies end up with?
Usually a connector for plumbing, some assistant seats, and one replaced operation. Consolidating onto a single vendor is a preference rather than a result.
How much should we budget?
Compare against the cost of the operation, not against the tool. The relevant number is what the two-day quote turnaround or the third support hire costs you per year.
Read next
If you are comparing options
Find out what your most expensive operation is
The assessment covers the whole company: how you get customers, how you deliver, and how the place runs. You get it in writing, and you can build from it with or without us.