Searching for developers with your own AI assistant: how it works

This page explains how that works in plain words, what an assistant can and can't judge, and how it works on Arca.ph.

What does "searching with your AI assistant" actually mean?

Today, most hiring tools give you a website to log into. You type filters, scroll profiles and copy notes into a spreadsheet.

With an assistant-based search, you skip most of that. You describe the role in normal sentences, for example: "Find a backend developer with Python and Postgres experience who can overlap at least four hours with our working day and wants pay within our budget." The assistant then asks the marketplace for matching profiles, reads them, and sorts them for you.

The assistant does the reading and sorting. You make the decisions.

What is MCP, in plain words?

To search a marketplace, your assistant needs a safe, standard way to plug into it. One common way is MCP, short for Model Context Protocol.

MCP is an open standard that lets AI apps connect to outside data and tools. The MCP project itself compares it to "a USB-C port for AI applications": one standard plug, so an assistant like Claude or ChatGPT can connect to many different systems without custom wiring for each one (modelcontextprotocol.io).

A company that wants assistants to reach its data publishes an MCP server. That's the "socket" on the company's side. You add the server's address to your assistant once, and from then on the assistant can use the tools that server offers, such as "search candidates" or "request an introduction."

You may also hear about an API (application programming interface). An API is a set of rules that lets one piece of software ask another for data. MCP is a standard way to offer that kind of access specifically to AI assistants.

Can an AI assistant really shortlist developers?

Yes, for the parts of hiring that are about reading and comparing information. An assistant is good at:

• Filtering on hard facts: skills, years of experience, desired pay, working hours.
• Ranking candidates against your job description and explaining why each one ranked where it did.
• Summarising long material such as resumes, screening answers and interview transcripts.
• Spotting gaps, for example "this candidate says they have strong English but has no English test on file."

An assistant is only as good as the data it can see. If a profile is a free-text resume and nothing else, the shortlist will just be a well-sorted pile of claims.

What can't an AI assistant decide for you?

Keep a human in charge of:

• Interviews. Talk to people before you hire them.
• Judging fit. Team culture, communication style and trust are your call.
• Offers and contracts. Pay, start date and terms are between you and the candidate.
• Contacting candidates. A good setup should ask for your confirmation before anything is sent, and the candidate should agree before their details are shared further.

Treat the assistant's shortlist as a first draft, not a verdict.

How does Arca.ph work with your assistant?

Arca.ph is a self-serve marketplace of vetted remote software engineers from around the world.

You can read every profile, or use an AI assistant you already have, like ChatGPT, Claude, or Grok, to search and sort them. You can connect that assistant to Arca.ph (arca.ph/how-it-works). It works like this:

1. Look at profiles for free. Every developer fills out the same profile: resume and work history, the tools they've worked with, the monthly pay they're asking for, their English level in their own words, the hours they can work, screening answers, a personality test and a written transcript of a video interview with an AI interviewer.
2. You or your assistant sort them. Arca.ph doesn't make the shortlist for you. You choose who you like.
3. Order proof if you want it. A real person watches each test: a coding test with no AI, a 2-hour build where AI is allowed, a live pair-programming session (the Fire Drill), an English test, a thinking-skills test and a typing test. Tests are optional and cost extra (arca.ph/proctored-tests).
4. Ask for introductions. Arca.ph contacts a developer only when you ask. If the developer says yes, that's an introduction. Then you run your own technical interview and decide.
5. Hire, and pay once. On the marketplace, if you hire someone Arca.ph introduced within 1 year of the introduction, you pay one fee of 15% of first-year pay, once. For example, a developer paid $3,000 a month earns $36,000 a year, so the fee is $5,400, once. That example is not a quote. If the hire doesn't work out, Arca.ph finds you a replacement for free within 3 months (arca.ph/pricing). After the hire, you arrange payroll and contracts with the developer yourself.

No AI assistant? You can browse and sort profiles yourself.

Is it safe to let an assistant search candidate profiles?

It can be, if the setup respects both sides. Look for:

• Your confirmation before any outreach. Nothing should be sent to a candidate or a vendor without you saying yes.
• Candidate consent. Candidates should agree before their contact details are shared or an introduction is made.
• Clear limits on what the assistant can do. Read-only search is low risk. Actions that send messages or place orders should always ask first.
• Your own company's AI policy. Check what your assistant is allowed to store or share, especially for personal data.

How do I get started?

1. Write a short job brief: skills, seniority, pay range and the working hours you need to overlap.
2. Decide which proof matters for the role (for example, a coding test with no AI for fundamentals, or a 2-hour build where AI is allowed if the job is building with AI tools).
3. Check whether the marketplace you're using connects to your assistant, and what the assistant is allowed to do there.
4. Let the assistant draft a shortlist, then interview the people on it yourself.

Get started with Arca.ph.