Questions we answerAI and automation
Do I need an AI agent, or just better tools?
Usually better tools first. An agent runs a defined workflow inside boundaries; a tool helps a person do their own work faster. The diagnostic tells you which the evidence supports.
Free · a few minutes · starts from “I think AI could help”
The short answer
Most teams want better tools. An agent — something that acts on its own, with a person holding the line — is justified where the workflow is repeatable, needs judgement along the way, comes round often, and where someone can say what happens if it acts wrongly. Where the judgement is high, a person with AI assistance fits better. Where the steps are fixed and mechanical, ordinary automation is cheaper and more reliable. Agents are the most expensive of the three and the easiest to reach for too early.
The words get used interchangeably and the decisions are not. Giving a person an AI tool is quick and reversible. Building an agent is a small project with oversight, permissions and escalation to design.
Many teams that ask for an agent turn out to need an assistant, and some turn out to need a feature of software they already pay for.
Is it really this?
Worth checking before you spend anything. These are the signals that tend to tell the two apart.
Signs it is
- The workflow runs end to end with the same steps and needs no one to decide anything mid-way.
- It happens often, and the delay between steps is itself a cost.
- The systems involved can be reached programmatically, and someone will own the agent when it misbehaves.
Signs it's something else
- A person is already doing the work well and is simply slow at the writing and checking parts.
- The steps vary, or the right next action depends on judgement.
- The software you already pay for has most of this built in and nobody has turned it on.
What's usually behind it
- A repetitive process that has never been simplified.
- Drafting and analysis work that a person could do faster with help.
- Software already in place that nobody has fully switched on.
What people usually get wrong
- Using the word agent for anything with a language model in it, and budgeting accordingly.
- Building an agent for a process that is neither written down nor stable.
- Giving an agent permissions nobody has reviewed, because it was quicker.
The options, and when each one fits
In no particular order. Which of these fits depends on your situation, not on the question you typed.
Better tools for the people you have
A person stays in charge and the work around their judgement gets faster.
Configure the software you already own
The capability exists in your current systems and has never been set up.
Build an agent for one workflow
The steps are stable, frequent, low-judgement and bounded, with a named owner.
Simplify the process first
The workflow is repetitive but nobody has written it down or removed the waste.
When doing nothing is the right call
If the process changes every few weeks, anything built now will be rebuilt. Waiting until it settles is not caution; it is the difference between building once and building three times.
What the diagnostic asks
- Where you think AI could help.
- How repeatable the work is, how much judgement it needs and how often it comes up.
- How sensitive the information is, what happens if it goes wrong, and whether there is software in place.
“I don't know” is always an answer. Nothing is assumed on your behalf.
The routes it weighs
Every diagnostic weighs all ten routes. These are the ones that tend to matter for this question — a description of the rules, not a promise about your answer.
AI Assist
Keep a person in charge and give them AI for drafting, summarising, checking or analysis.
AI Agent
Delegate a defined, repeatable workflow to software inside boundaries you set, with human oversight.
Process improvement
Change how the work flows — remove steps, name an owner, fix the hand-offs.
What changes the answer
- Judgement-heavy or high-stakes work points at a person with better tools, not an agent.
- An existing system that already covers most of the process may be the whole answer.
Get your own read
A handful of short questions, then the strongest route for your situation, what's also credible, what we still don't know and what would change our view. No account, no setup.
Related questions
- Should I automate this work?If it is repeatable, frequent, low-judgement and written down, quite possibly. If any of those is missing, fix that first.
- Repetitive admin is eating the weekWork out how much of it should exist at all before automating any of it. Some repetitive admin is a process nobody has simplified, some is a good candidate for software, and a little of it genuinely needs a person.
- Can AI help my team?Often, but the useful question is where. AI that helps a person draft, summarise or analyse is a different decision from AI that runs a process.
- Can AI handle customer enquiries?Part of them, well, with a person holding the line. The split that matters is between enquiries with known answers and those needing judgement, empathy or authority.
- Is my process ready to automate?Four things have to be true: the steps repeat, they are written down, it happens often enough to pay back, and someone can say what a failure costs.
- Should I automate reporting?Usually yes, and usually after asking which reports anyone actually uses. Automating a report nobody reads makes it permanent.