AI + VAs

AI vs a virtual assistant: which does your business actually need?

By Jenn and the DotVA team. We run an Australian VA agency on Claude Code and ChatGPT every day, so this is lived experience, not theory. Last updated 09/06/2026.

The question we get most often in 2026 is some version of "do I still need a VA if I have AI?" It is the wrong question. AI and a virtual assistant are not competing for the same job. AI is a brilliant, tireless, cheap producer of outputs, and a genuinely poor owner of outcomes. A good VA is the opposite: slower and more expensive per task, but able to hold context, exercise judgement, talk to your customers and be accountable when something is wrong.

Put the two together and the maths changes. The repeatable 60-70% of most support work (drafting, summarising, reformatting, first-pass research, data entry) collapses to minutes with AI, which frees the VA to spend their hours on the judgement work that AI cannot be trusted with. That is the whole idea behind an AI-augmented VA, and it is how every DotVA placement is set up to work.

When to reach for each

Reach for AI when

  • The task is high-volume and low-stakes
  • A human will check the output before it goes anywhere
  • You need a fast first draft, a summary, or a reformat
  • You are exploring or researching a starting point
  • Being slightly wrong costs you nothing

Reach for a VA when

  • The task needs judgement, context or relationships
  • It touches a customer, supplier or your reputation
  • Someone has to own the result and be accountable
  • It requires chasing, deciding, or saying no
  • Being wrong actually costs you money or trust

Use both (the AI-augmented VA) when

  • It is real work, which is most of the time
  • The VA drives and stays the accountable owner
  • AI accelerates the repeatable parts of the task
  • A guardrail check sits between AI speed and your inbox
  • You want more done per hour, not a riskier shortcut

The playbooks: how we actually do it

Everything below is written from inside DotVA, by the team running these workflows in production. Read it as a methodology, not a blog feed: start with the philosophy, pick your tools, run the workflows, then train and guardrail so AI speed never turns into AI error.

Start with the philosophy

Why we pair AI with VAs instead of betting on one. The honest case, and the numbers behind it.

  1. Why we don't replace VAs with AI (and probably never will)

    The contrarian take. AI is brilliant at outputs and bad at judgement. The seven tasks where AI consistently loses to a competent VA.

    Read the playbook →

  2. The ROI of an AI-augmented VA: real numbers from a Melbourne agency

    What 12 months of Claude Pro + ChatGPT Plus actually saves a VA in hours. Spreadsheet downloadable.

    Read the playbook →

Pick the tools

What we actually pay for, what we dropped, and which tool wins for which job.

  1. Claude vs ChatGPT for virtual assistants: which one wins in 2026

    Side-by-side, 14 real tasks. What each is good at, what they break on, what we actually pay for.

    Read the playbook →

  2. The AI-augmented VA stack: what to pay for in 2026

    Claude Pro vs ChatGPT Plus vs Perplexity vs your VA. Where the marginal dollar wins.

    Read the playbook →

  3. Claude Code for non-developers: 5 commands every VA should know

    Five Claude Code commands that work for non-coders. Each with a real use case from our team.

    Read the playbook →

Run the workflows

The day-to-day playbooks: the exact ways our VAs put AI to work without dropping the ball.

  1. How we run our VA team on Claude Code (a working playbook)

    Claude Code is usually pitched as a developer tool. We use it as a VA productivity layer instead. Five daily workflows and the failure modes.

    Read the playbook →

  2. Shopify automation for VAs: 12 tasks your VA can hand to Claude

    The exact prompts our team uses to run e-commerce operations with VAs + Claude Code.

    Read the playbook →

  3. Writing SOPs your VA AND Claude can both follow

    A 5-step template for SOPs that work for humans and AI. Examples included.

    Read the playbook →

  4. Building a Telegram briefing bot for your VA in 30 minutes

    A free morning briefing bot that wakes your VA up with the metrics that matter. No code required.

    Read the playbook →

Train and guardrail

How a new VA learns the AI stack in week one, and the guardrails that stop a confident-but-wrong answer reaching you.

  1. Training a new VA on AI in their first week: the DotVA checklist

    Our actual onboarding doc, redacted and shared. Every AI workflow your VA needs in week 1.

    Read the playbook →

  2. AI hallucination guardrails: what we built into our VA workflow

    The four guardrails we put around every AI-augmented task to stop a VA from shipping a hallucination.

    Read the playbook →

  3. Prompt library for Australian VAs: 40 ready-to-use prompts

    Admin, content, research, customer support. Australian English, AU context, drop-in ready.

    Read the playbook →

AI and virtual assistants: common questions

Will AI replace virtual assistants?

Not for the work that actually matters. AI is excellent at producing outputs (drafts, summaries, reformatting, first-pass research) but unreliable at judgement, context and accountability. It does not know your customers, cannot be trusted to decide what to send, and will state a wrong answer with total confidence. The businesses winning with AI are not replacing their VA, they are giving a capable VA AI as a power tool, which is faster and far safer than either alone.

What is an AI-augmented virtual assistant?

An AI-augmented VA is a trained human assistant who uses AI tools (Claude, ChatGPT and similar) to do more, faster, while staying the accountable owner of the result. The VA drives, AI accelerates the repeatable parts, and the VA checks and ships. At DotVA every VA is trained on this stack in their first week, with guardrails so AI speed never becomes AI error in your inbox.

Can a virtual assistant use AI safely with my customer data?

Yes, with the right guardrails, which is exactly where most casual AI use goes wrong. We work to a least-privilege principle (the VA only accesses what the task genuinely needs), keep sensitive data out of public AI tools, and run a four-part hallucination-guardrail check on AI-assisted work before anything reaches you or a customer. You also remain accountable for personal information you share under the Australian Privacy Act, so a managed process matters.

Does an AI-augmented VA cost more than a normal VA?

No. DotVA rates are the same: admin VAs $12-17/hr, specialists $18-25/hr and bookkeepers $25-35/hr (AUD, excl GST). The AI training and tooling are part of how we work, not a surcharge. The point of the model is that you get more done per hour, not that you pay more per hour.

Which AI tools do your VAs actually use?

Mostly Claude and ChatGPT, with Claude Code as a productivity layer for non-developers, plus Perplexity for research and a few task-specific tools. We have written up the exact stack, what each is best at, and where the marginal dollar is worth spending, rather than paying for everything.

Free toolkit

Get our AI-augmented VA toolkit

The 40-prompt library our VAs use daily, plus the SOP template that makes a task work for a human and Claude both. Sent to your inbox, free.