AUSPEXAW-speks
Agentic Engineering Diagnostic
Auspex is Latin for the official who read the signs before any undertaking. Our Diagnostic reads them in code, before you invest in a build.
Bring one workflow. Leave with the code, a blueprint, and a clear next step.
Is this for you?
- Owners exploring AI for one business workflow
- Teams who want code to inspect and an engineer to discuss it with
- Businesses deciding what to build or evaluate next
The situation
An idea for AI is easier to discuss than to judge.
You have a workflow that looks like a good candidate for AI. It may be a report someone assembles by hand, a data clean-up that never quite ends, or a tool your team keeps asking for. The conversation about it has gone as far as conversation can.
What you need now is something to look at. Slides and estimates describe what might happen; code shows how the work actually behaves. It surfaces the awkward data, the missing rule, and the step nobody wrote down.
You also want an engineer’s view before committing a larger budget. The question is practical: is this worth building, and what would the next step cost?
Our approach
Real work first, then a senior conversation about it.
The Diagnostic starts with one bounded coding task that we agree with you. An agent works on it for one hour under an engineer’s direction, and the engineer reviews what it produces before anything is delivered.
You receive the code in a private repository, or as a ZIP if you prefer. The README explains how to run it, what the hour produced, and what work remains. Nothing is deployed to your systems; you decide what, if anything, runs next.
The one-hour consultation then works from that code. We discuss where AI applies to the workflow, the risks we saw, where people need to approve, and the next step that fits. The fee is fixed at $750 USD, so the cost of the answer is known before you start.
Use cases by industry
Where this service fits.
Typical applications across industries. They show where the service applies, not past client work or results.
- Accounting firm
Client document intake sorted for review
An agent writes code that reads sample client uploads and sorts them by document type and tax year. Uncertain files are flagged rather than filed. The consultation covers naming conventions, the document types the samples missed, and where a staff accountant confirms each sort before filing.
- Distributor
Purchase order lines matched to supplier catalogs
The hour produces code that matches sample purchase order lines to a supplier catalog export. Unmatched lines are listed for a buyer to resolve. The consultation covers unit conversions, price changes, and substitutions, and where the purchasing lead approves each order before it is sent.
- Property management
Maintenance requests triaged by urgency and trade
An agent drafts code that reads sample tenant requests and proposes an urgency level and trade for each. The property manager reviews every proposal before a vendor is dispatched. The consultation looks at the edge cases the samples surfaced and what a fuller build would need from the maintenance system.
- Dental group
Insurance verification notes from de-identified forms
Working from de-identified sample forms, the hour produces code that extracts coverage details into a standard verification note. Front-desk staff confirm each note before it is relied on. The consultation covers privacy obligations, the gaps the samples showed, and how the practice management system might receive the notes.
- Marketing agency
Campaign summaries drafted from platform exports
An agent writes code that combines sample advertising exports into one draft summary per client. An account manager edits and approves each summary before it is sent. The consultation weighs how often the export formats change and what keeping the code current would ask of the team.
- E-commerce
Product listings checked against catalog rules
The hour produces code that checks sample listings for missing attributes, inconsistent sizing, and rule conflicts. It prepares suggested fixes for the merchandiser to accept or reject. The consultation separates rules that belong in code from calls that need a merchandiser’s judgment.
- Field services
Technician notes turned into structured job reports
An agent writes code that turns sample technician notes into job reports listing parts, labor, and follow-up items. The service coordinator reviews each report before it goes to billing. The consultation covers missing fields, note quality, and the next step toward the field service system.
- Nonprofit
Funder report fields assembled from program records
The hour produces code that gathers sample program records into the fields a funder report requires. Gaps and conflicting figures are flagged for the program director. The consultation covers which records are reliable enough to use and where the director approves each report before submission.
What you receive
One hour of agent coding
An agent works on the agreed workflow for one hour. You receive the code produced for review.
An Agentic AI Blueprint
A one-page plan for the workflow: where agents help, where people approve, the data boundary, a first measure, and the next step.
One hour of consultation
Discuss the code, the blueprint, AI applicability, risks, and next steps with an engineer.
A clear next step
Leave knowing where AI fits the workflow and what further work it needs. People approve any use or deployment.
How it works
Describe the workflow
Share the task, systems, and intended outcome. Leave confidential records and credentials out of the request.
Agree the coding task
Choose a bounded task for the one hour of agent coding and identify the human approval points.
Receive the code
Receive a private repository, or a ZIP if you do not use GitHub. A short README explains setup, work completed, and remaining work.
Consult and decide
Review the code and your Agentic AI Blueprint in the one-hour consultation, then decide your next step. You approve further work and deployment.
How the Diagnostic is checked
What the hour and the consultation deliver.
Each item can be checked against the code, the README, or the consultation.
- Task agreed in writing
- The bounded task and its approval points are written down before the hour starts. The README records them, so the result can be checked against the agreed task.
- Code runs as documented
- The README’s setup steps are followed on the delivered code. It runs on the sample data as the README describes.
- Completed and remaining work
- The README separates what the hour produced from what remains. Each item can be checked against the code itself.
- Approval points named
- Each place a person reviews or approves is named and discussed in the consultation. The workflow owner confirms they match how the work runs.
- A next step you can price
- The consultation ends with one named next step and what it would involve. The workflow owner confirms it is clear enough for a budget decision.
Where care is needed
What we watch, and how it is handled.
- Confidential data
- The hour runs on representative samples or descriptions that contain no confidential records. Any access to sensitive material is agreed with the engineer before it is shared.
- Task size
- An hour of agent coding suits a task with a clear input and a clear output. We shape the task with you so the hour produces something worth reviewing.
- Agent-written code
- An engineer directs the agent and reviews its code before delivery. The README states what was reviewed and what still needs testing.
- Partial results
- An unfinished task can still answer the main question about the workflow. The consultation explains what the code shows and what it leaves open.
- What runs next
- The code is delivered to you in a private repository or as a ZIP. Nothing is deployed to your systems, and any further use is your decision.
Who does what
Your team decides. We engineer.
Your team
- Choose one workflow and describe its intended outcome
- Agree the bounded task for the hour of agent coding
- Provide non-confidential samples or descriptions of the data
- Bring the workflow owner to the consultation
- Decide and approve any further work or deployment
Sophrono
- Help shape a task that fits one hour of agent coding
- Direct the agent and review the code it produces
- Deliver the code with a README on setup, output, and remaining work
- Lead the consultation on applicability, risks, and approval points
- Recommend a next step, including when AI is not the right fit
At the end
The decisions you make next.
The service ends with evidence and a choice. Each option is yours, and none is assumed.
Plan the larger system
When the workflow belongs to something bigger, a System Blueprint designs it and prices the build against an Acceptance Charter you sign.
Test it under production conditions
When the workflow is bounded and clearly owned, a Production Pilot runs it with real integrations against criteria you approve.
Continue with your own team
Your developers carry the code forward from the README. Its setup and remaining-work notes are written for that purpose.
Pause here
If AI is not the right fit yet, you pause with a clear reason. The code and README stay with you for later.
Before we start
What to have ready.
- A short description of the workflow and the outcome you want
- The systems or files the workflow reads from and writes to
- Representative examples that contain no confidential records
- The person who owns the workflow and can decide the next step
- A GitHub account for the private repository, or a preference for a ZIP
Related services
What arrives with the code
A README anyone on your team can follow.
Every Diagnostic delivery includes a README with the same six parts. This is the outline; yours is written for your task. People review the code and approve any use or deployment.
Agreed task
- Task
- The bounded task chosen for the hour of agent coding
- Version
- The commit or archive you received
- Consultation
- When the one-hour review takes place
What the hour produced
- Delivered
- What runs, what is scaffolding, and what is unfinished, stated separately
- Blueprint
- Where to find your Agentic AI Blueprint, the one-page plan delivered with this code
Run locally
- Setup
- Runtime, install command, and configuration names, never secret values
- Try it
- A safe example input and the result to expect
Review and evidence
- Checks run
- Each check, where it ran, and its result
- Not tested
- What was left out, named plainly
- Approval points
- Actions that need a person, and whose role that is
Remaining work
- Next
- Implementation, evaluation, integration, and deployment work still to do
Dependencies and data
- Inventory
- Libraries and their licenses, plus any approved data requirements
Auspex · Agentic Engineering Diagnostic · $750 USD
Bring one workflow. Leave with the code, a blueprint, and a clear next step.
One hour of agent coding, the code delivered to you, an Agentic AI Blueprint, and a one-hour consultation.
What the Diagnostic includes
- One hour of agent coding for the agreed task
- The code produced, delivered to you
- An Agentic AI Blueprint: a one-page plan for the workflow
- One hour of consultation on the work, the blueprint, and next steps
Receive a private repository, or a ZIP if you do not use GitHub. A short README covers setup, completed work, and what remains before production.
People review the code and approve any use or deployment. Full engagement terms are finalized in a Master Services Agreement.
What “accepted” means
Measured against criteria you agree to in advance.
- One hour of agent coding is completed for the agreed task.
- You receive the code produced during that hour.
- You receive an Agentic AI Blueprint for the workflow.
- A one-hour consultation covers the work, the blueprint, AI applicability, risks, and next steps.
The keystone
Architecture is decisive.
As implementation becomes faster and more accessible, the quality of the architecture increasingly determines the outcome. Agentic Engineering Diagnostic is one of the ways we hold to it.
Read Canon IFree worksheet
The four questions we answer before we build.
Use the same four questions 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.
- Which parts should agents perform?
- Which decisions stay with senior people?
- How is every output checked automatically?
- How is parallel work reconciled?
The Four Questions worksheet
Preview it firstWhere this fits
- Learn Research
- Explore free AI Workload Evaluation
- Try one workflow Agentic Engineering Diagnostic
- Plan and validate System Blueprint
- Build Custom Agentic Systems
- Improve and operate AI Stewardship
Not ready yet? The Four Questions. After this: Production Pilot or System Blueprint.
Questions
What is included in the fee?
The $750 USD Diagnostic includes one hour of agent coding, delivery of that code, an Agentic AI Blueprint, and one hour of consultation.
Is “Ask an engineer” the same thing?
Ask an engineer is a free 15-minute call in which an engineer names the agentic systems that could fit your workflow. The Diagnostic adds the code, the full Agentic AI Blueprint, and a one-hour consultation.
Do we get a written plan?
Yes. The Agentic AI Blueprint is a one-page plan for the workflow, delivered with the code. The README records what the hour produced and what remains.
Is the Agentic AI Blueprint the same as the System Blueprint?
No. The Agentic AI Blueprint plans one workflow on one page and comes with the Diagnostic. The System Blueprint designs a whole system and its acceptance criteria, as a separate scoped engagement.
How is the code delivered?
We provide a private repository, with a ZIP alternative. The README covers how to run it, what the hour produced, and remaining work.
What should we send?
Describe the workflow, systems, and intended outcome. Agree any data access with the engineer before sharing confidential material.
What if AI does not fit the workflow?
We say so plainly in the consultation, and discuss the next steps that do fit the work.
How is the Diagnostic paid for?
You request it here, and we reply with an engagement letter for the task. Payment is arranged directly with you after you sign it, by invoice if you need one. No payment is taken on the website.
What governs further work?
Full engagement terms are finalized in a Master Services Agreement.
Will the hour finish the task?
Not necessarily, and the fee does not depend on it. The hour produces real work to review, and the README records what remains.
Is the code ready for production?
It is written for review and discussion. Production use needs further work, which the consultation helps you scope.
Can our own developers continue the work?
Yes. The README explains setup and remaining work, so your team can run the code and carry it forward.
Who should join the consultation?
The workflow owner, and anyone who would maintain or approve the code. The discussion goes further when the decision-maker is present.