How we use AI
Engineer Austin, a licensed engineering firm in Georgetown, Texas, uses AI assistants inside its IT and engineering workflow to draft, analyze and cross-check, under rules that keep every decision and every change with a person. Below: where AI is used, the guardrails, what clients notice, and how we help organizations adopt AI with the same discipline.
Engineering is mostly careful repetition. That is what AI is good at.
A network assessment means reading thousands of log lines and comparing dozens of configurations against a written standard. A migration means drafting scripts, checking them, running them and writing down what happened. A report means turning all of it into findings a board can act on. The judgment in that work belongs to people: our team's, with our engineer brought in when a question needs a licensed engineer. Most of the hours are not judgment.
We use AI assistants, large language models operated through approval-gated tooling (nothing they propose runs until a person approves it), to take the first pass at that repetition: drafting and checking configurations and scripts, analyzing logs and inventories, writing and cross-checking reports and runbooks, building internal software, and researching vendor documentation. Our team still measures, decides and signs off, and our engineer is brought in when a question needs a licensed engineer. The assistant types, compares and summarizes, and it does so under rules we wrote before we let it near a client system.
We treat AI the way we treat any new platform: adopted when it earns its place, governed by a written standard, re-evaluated on a schedule, and described to clients plainly. This page is part of that standard.
In one paragraph
AI assistants help us work faster and more thoroughly. Every AI-assisted change is reviewed by a person, logged with a timestamp and a rollback, and approved before it runs. Client secrets are never pasted into public AI tools; work is done in controlled sessions with access limited to what the task needs. A named member of our team reviews and signs off on the result.
Where AI is used in our work
Five kinds of work, each with a person at the end of it.
- Drafting and checking configurations and scripts. Firewall rule sets, router builds, Group Policy, PowerShell inventory and remediation scripts are drafted against our house standards and checked for drift before a member of our staff reviews the diff.
- Analyzing logs and inventories. Security logs, event logs, device inventories and carrier call records are sorted, correlated and summarized so the person doing the work starts from a short list of what matters instead of a long list of everything.
- Writing and cross-checking reports and runbooks. Findings registers, risk registers, restore procedures and as-built documents are drafted from the record and cross-checked against the evidence, then edited and signed by a person.
- Building internal software. Our own service-request, quoting and documentation tools are built faster, with tests and review gates between the draft and the release.
- Researching vendor documentation. Release notes, permission requirements and upgrade paths are gathered and compared, and the answer is verified against the vendor's own source before it is relied on.
Not used for: deciding what you should buy, judging what is proportionate to your risk, sealing engineering work, or changing anything on your systems without a person's review and your written approval.
The rules that do not bend
These are the same change-control rules we apply to every firewall, server and tenant, applied to a new kind of helper. None of them is optional.
A person reviews every change
An AI-drafted configuration, script or document is a draft until a qualified person has read it. Nothing moves from draft to applied without that review.
Logged with a timestamp and a rollback
Every AI-assisted change is recorded the way a hand-written one is: a timestamped log of what ran, and a rollback script written alongside, so the way back exists before the step forward.
Approved before it runs
The tooling is approval-gated. Each proposed action is shown to the person doing the work and approved before it executes. Your written approval for the work came first, as it does for any change.
No client secrets in public AI tools
Passwords, keys, tokens and client data are never pasted into public AI services. The assistants we use run in controlled sessions, with access limited to what the task in front of them needs.
A person reviews and signs off
The report, the design and the change carry the name of the person on our team who did the work, held to standards set by our licensed Professional Engineer. AI does not replace that person's judgment, and it does not share their accountability.
Re-evaluated on a schedule
The tools, the rules and this page are reviewed as the field changes, on the same cycle as the rest of our standards. When the policy changes, clients hear what changed and why.
What clients notice
You do not see the assistant. You see its effect on the work, in four places.
- Faster turnaround on assessments, reports and migrations. The analysis that used to take days of reading takes hours, and our team's time goes to confirming and deciding.
- More thorough documentation at the same price. Findings registers are longer and better cross-referenced, runbooks stay current, and as-builts are written at the time rather than reconstructed later.
- Staff spending their time on judgment, not typing. What is proportionate, what to retire, what the risk really is: the questions only a person can answer get more of the person.
- The same standards applied every time, because the checklists are executed rather than remembered. A step cannot be skipped because it was Friday afternoon.
What does not change
- The price of the program; AI assistance is how we work, not a line item
- Who you talk to; a person answers, decides and signs
- Your written approval before anything is bought or changed
- Your ownership of every record, design and credential
- The two-hour business-hours response target
What we offer organizations adopting AI
The same questions we answered for ourselves, answered for you: what is it good for, what must be true before it touches your data, and who is accountable. Advisory, scoped and delivered under the same assess-first, written-approval process as everything else.
AI use policy and governance
A written policy for your organization: which tools staff may use, what data may never leave the building, who approves exceptions, how use is logged. Sized to a two-person office or a county department, and aligned with your records-retention and public-information obligations where they apply.
Microsoft 365 Copilot evaluation and rollout
Whether Copilot pays off for your users, and for which ones, before anyone is licensed. Licensing through our Microsoft Cloud Solution Provider relationship, billed monthly and owned by you, with the tenant settings that keep it from reaching what it should not.
Data readiness
What must be cleaned, labeled, permissioned or locked down before an AI tool is allowed to read it. In Microsoft 365 this usually means SharePoint structure, sharing policy and sensitivity labels; we fix those first, because an assistant that can see everything will.
Automation and custom software with AI components
Internal tools that use AI for classification, extraction or drafting, with a human review step designed in from the start. Built by our software team with tests and review gates, and owned by you, code included.
Staff training on safe use
Short, practical sessions for the people who will actually use the tools: what to paste and what never to paste, how to check an answer, when to stop and ask. Paired with the state-required cybersecurity training for Texas public entities where it applies.
Where to start
With an assessment. We look at what you have, what your staff already do with AI tools (usually more than anyone has written down), and where the risk and the value sit. You get a written report and a recommendation; you decide. See Cloud and Microsoft 365 and IT consulting.
What we do not claim
We do not offer a helpdesk run by AI, AI monitoring of your network, or anything that changes your systems on its own. The assistants are tools we use, not a product we sell, and we do not dress them up as one. We do not let a tool make an engineering judgment, and we do not pretend one could. AI makes a careful firm faster. It does not make a careless one careful, and it is not a substitute for the people whose names are on the work.
AI at Engineer Austin, answered plainly
Is our data used to train AI models?
Not by us. Client data, passwords, keys and tokens are never pasted into public AI tools, where inputs may be retained. The assistants we use run in controlled sessions through approval-gated tooling, with access limited to what the task needs, and the lasting record of the work is the log, the change and the report, which we keep under document control and you own.
Who is accountable for AI-assisted work?
People, never the tool. The person on our team who does the work reviews and signs off on it, the same as for any other change. An AI assistant has no authority in our process. It drafts, compares and summarizes; a person reviews, approves and signs. If an AI-assisted change were wrong, it would be our error, corrected in writing like any other.
Can AI change our systems on its own?
No. Nothing changes on a client system without your written approval and a person's review. The tooling is approval-gated: every proposed change is shown to the person doing the work, logged with a timestamp and a rollback, and approved before it runs. There is no autonomous agent watching or altering your network, and we do not sell one.
What does this cost us?
Nothing extra. AI assistance is part of how we work, like our scripts and checklists; it is not a line item or a tier. You see the benefit as faster turnaround and more thorough documentation at the same price. Advisory work on AI readiness for your own organization is a separate, scoped engagement, quoted in writing like any project.
Can you help us adopt Microsoft 365 Copilot safely?
Yes. We start with data readiness: what Copilot would be able to reach in your tenant, what should be cleaned, labeled or locked down first, and which users would get value from a license. Copilot licensing is available through our Microsoft Cloud Solution Provider relationship and remains your property. Rollout follows the same assess-first, written-approval process as everything else we do.
What is your AI use policy?
In short: AI assistants may draft, analyze and check; they may not decide or act. Every AI-assisted change is reviewed by a person, logged with a timestamp and a rollback, and approved before it runs. Client secrets are never pasted into public AI tools; sessions are controlled and access is limited to the task. A named member of our team reviews and signs off on the result. We revise the policy as the tools change, and we tell clients what changed.
Start with a written assessment.
Tell us what you have and what is not working. Someone from our team, not a salesperson, replies within one business day by email or phone, and brings in our engineer when your request needs one. The assessment report is yours to keep whether or not you buy anything.