Free worksheet · Our method
The four questions we answer before we build.
Use them on your own workflow before you talk to any developer. It takes about twenty minutes, and it makes every proposal you receive easier to compare.
- The four questions, with prompts for each
- A step-by-step scoring table for your workflow
- The same method we use in the Auspex Diagnostic
Get the four questions worksheet free
Preview
The questions, and why they matter.
The full worksheet adds every prompt, space to write, and the scoring table.
Which parts of the system should agents perform?
Agents pay off on work that is frequent, rules-based, and checkable. They cost more than they save on rare, ambiguous, or high-stakes work.
- List every step of the workflow, from the moment work arrives to the moment it’s done.
- For each step: how often does it happen each week, and how long does it take?
Which decisions remain under senior control?
Authority stays human. The question is where exactly the line sits, and who holds it.
- Which decisions commit money, or change what a customer is promised?
- Which decisions change a record of truth, such as a ledger, a contract, or a patient or client file?
How is every agent’s output validated automatically?
If you can’t check an output automatically, a person has to check all of it, and most of the savings disappear.
- What does a correct result look like? Write down two good examples and two bad ones.
- Can correctness be checked against a rule, a record, or a total?
How is parallel work reconciled within the architecture?
Several agents working at once is fast only if their work fits back together cleanly.
- Which system holds the truth for each kind of record?
- What happens if two agents, or an agent and a person, change the same record?
Who it is for
Anyone deciding where agents belong in a workflow.
- You have one workflow in mind and want to know which steps suit agents.
- You expect proposals from developers and want a fair way to compare them.
- You need to decide who keeps authority over money, records, and promises to customers.
- You want your own answers written down before anyone writes code.
What is inside
Four questions, their prompts, and a scoring table.
- Which parts should agents perform?Prompts on each step’s frequency, duration, rules, and the cost of a mistake.
- Which decisions remain under senior control?Prompts on decisions that commit money, change a record, cannot be undone, or need professional judgment.
- How is every output validated automatically?Prompts on good and bad examples, checks against rules or records, and who reviews exceptions.
- How is parallel work reconciled?Prompts on which system holds the truth and who decides when outputs conflict.
Each question explains why it matters and gives four prompts with space to write. The scoring table rates each workflow step on volume, rules, checkability, the cost of an error, and who approves.
How to use it
One workflow, start to finish.
- Choose one workflowPick a single process, from the moment work arrives to the moment it is done.
- List and score every stepFill in one row of the scoring table per step, using honest estimates.
- Mark the approval pointsName the person who approves each decision that commits money, changes a record, or cannot be undone.
- Write examples of correct workRecord two good results and two bad ones, so a check can be designed.
- Settle the system of recordNote which system holds the truth for each record, and who resolves conflicts.
Steps with high volume, clear rules, and checkable results are the candidates. Everything else stays with your team for now.
What to do next
Bring your answers to an engineer.
The worksheet frames the questions. The Diagnostic answers them on your workflow.
After the resource
Run the four questions with an engineer.
Questions
Do I need technical knowledge to use it?
No. The prompts ask about your workflow, your decisions, and your records. The person who knows the work best is the right person to fill it in.
What if most steps score low?
That is a useful answer. It may mean agents are not the right fit for that workflow yet, or that one step is worth testing on its own.
Does a high score mean we should build?
No. A high score marks a candidate. The decision comes after the test, the threshold, and the approver are agreed in writing.
Is this the same method Sophrono uses?
Yes. The Agentic Engineering Diagnostic starts from the same four questions, applied to one workflow with an engineer.