Sydney, Australia · AI agents for businesses and products
AI agent developer in Sydney.
VAI Soft Lab is a small software studio in Sydney, Australia, run by Vahid Alamipour, a lead engineer with 16+ years of building software, from junior developer to CTO. We design and build AI agents: software that reads, decides and acts on a real job, inside your business or your product. Our stack is Claude and OpenAI models, MCP, TypeScript, Python, Node.js, React, PostgreSQL and AWS. How we work: you send five lines about the idea, we reply with a plain answer, then we build a small first version on your real work, with people approving the important steps.
Real problems with AI agents
The demo works. Real work is harder.
An agent that looks good in a demo often fails on real emails, real files and real customers. These are the problems we plan for from the first day.
- It works on the demo, not on real data.Real emails are messy, forms are scanned crooked, names are spelled three ways. We test on a sample of your real work before anyone calls it done.
- Nobody knows when it is wrong.An agent can sound sure and still be wrong. We add checks in code, log every step, and send anything unclear to a person instead of guessing.
- Too much power.An agent with access to every system can do real damage. We give it the fewest tools it needs, read-only where we can, and a person approves anything that sends, pays or deletes.
- Hidden instructions.Words inside an email or a web page can try to steer an agent (prompt injection). We treat outside text as untrusted and keep the risky actions behind approval.
- Costs that grow quietly.Model calls add up, and multi-agent setups use many more tokens than a chat. We pick the smallest setup that does the job and track the cost per task.
- It does not fit your tools.The agent must work with your email, your files, your database and your apps. We connect it through their APIs or MCP, and keep your data in your accounts.
How it works
Four steps.
You decide after each step. Nothing starts before you agree.
- Send five linesWhat the job is, who does it today, how often, which tools it touches, and what a good result looks like. We reply within one working day with a plain answer: is an agent a good fit, a simpler automation, or not worth it.
- Agree the scopeIf it is a fit, we agree together what the first version does, what it must never do, and how we measure it. Then you get the time and the price in writing.
- Build a small first versionOne job, end to end, tested on your real work. People approve the important steps. Our own AI agent team does the heavy work; Vahid designs it, leads it and checks every change.
- Measure, then growWe look at the numbers together: time saved, mistakes caught, cost per task. Then you decide to stop, change it, or add the next job.
- Agents that read customer emails and forms and draft the reply or the quote.
- Agents that sort documents and type the data into your system.
- Search over your own documents, with the source shown.
- AI agents inside your product, for your users.
- Coding agent setups for development teams (Claude Code for teams).
- Let an agent move money or make big decisions about people on its own.
- Put your customers' data into a consumer AI tool.
- Promise numbers before we have tested on your real work.
Questions about AI agents
Questions.
What is the difference between an AI agent, a chatbot and an automation?
A chatbot answers questions. An automation follows fixed steps. An AI agent can read, decide the next step and use tools to do a job. Many jobs need only an automation, and we tell you when. More in our article on chatbot, automation or AI agent.
How do I know if an agent is worth it for my business?
Count how often the job happens and how long it takes a person today. If it is frequent, follows clear rules and costs real hours, it is a good first candidate. Our article on whether an AI agent will save your business time has a simple way to check.
Is my data safe? We are an Australian business.
We use business plans and APIs where your data is not used to train the models, keep your data in your own accounts where we can, and give the agent only the access it needs. If your business has privacy duties under Australian law, we design for them from the start.
Which AI model do you use?
The one that fits the job. Mostly Claude and OpenAI models, sometimes a smaller or local model when cost or privacy matters. We can switch models later without rebuilding the agent.
Can the agent work with the tools we already use?
Usually yes, if the tool has an API or an MCP server: email, calendars, files, databases, accounting and CRM apps. We check your tools in the first step, before any promise.
Do you only work with businesses in Sydney?
No. We are in Sydney, Australia, and work with clients in Australia and around the world, in English, by email and video call.
Next step
Send five lines about your idea.
Email hello@vaisoftlab.com with five lines: the job, who does it today, how often, which tools it touches, and what a good result looks like. You get a plain answer within one working day.
- Vahid Alamipour: 16+ years building software, from junior developer to CTO, then lead engineer roles in Malaysia and Australia.
- Has led 30+ developers into AI engineering. Over a year of daily work with coding agents.
- Four Anthropic certificates in building with Claude.
- Runs his own team of AI agents every day in VShell, the Mac app he built.
- Other ways to work with us: all services.