Sunday night, 9pm. There's a client starting Wednesday with nobody on the schedule, and you want to know one thing: is there anyone in the pipeline who can do weekend mornings and has done dementia care? You could log in, open the job and click through 38 applications. Or you could just ask.
That's what First Shift's connection to Claude and ChatGPT is for. You connect your account, and then you can ask your pipeline questions in plain English, from the same chat window you'd use for anything else.
What you can actually ask it
Under the hood it's an MCP server, which is just the standard way Claude and ChatGPT talk to other software. What matters is what it can see. It can list your jobs and how many applications each one has, open a job to see what it asks applicants, and show you the stages in your pipeline.
It can list the applications on a job (everyone still in New, everyone who applied since Friday), search across everyone in plain English ("caregivers with dementia experience who can do weekends"), and open one applicant: contact details, availability, their answers to your screening questions, and any stored AI analysis, including the fit score. It can pull an applicant's CV text when you need it. And it can line up two to eight applicants side by side.
Ask it what you'd ask a sharp assistant who'd actually read every application. "Who are the five strongest applicants for the weekend HHA role, and why?" "Compare these three for the Plano client." "Who's applied since Friday and is still sitting in New?"
There are three ready-made prompts built in as well: a shortlist for a job, a deep look at one applicant, and a head-to-head comparison. They sort people into strong, maybe and no, with the evidence, the gaps and what to check on the call, so you don't have to phrase the question perfectly every time.
What it can't do
It's read-only. It can't move anyone to a new stage, add a note, reject anyone or send an email. That's on purpose. When you've decided who to call, you go into First Shift and do it yourself.
It only knows what the application says. If a caregiver didn't mention CPR, the answer is "not stated", not "no CPR". It's told to keep those two apart, to cite the field behind every claim, and to say plainly when there's too little to judge. Not stated is a question for the phone call, not a reason to skip someone.
A missing fit score means nobody's scored that applicant yet. It doesn't mean a low score, and it's told not to treat it like one.
It won't search on protected characteristics. Ask it for applicants of a certain age, sex or religion, or anything along those lines, and the search is refused before a single application is read. It's also told not to guess at those things from a name or a CV.
And it doesn't make the hiring decision. Anything that leans on First Shift's AI analysis comes labelled as advisory, generated from what the applicant wrote, with the decision left to you. It's a faster way to read the pile. It isn't a verdict.
A worked example
Say you're a 22-caregiver agency in Plano with a weekend HHA ad up, and 38 applications on it. Made-up agency, realistic numbers. You ask: "Who on the weekend HHA job says they can do Saturday and Sunday mornings, and what have they told us about CPR and transport?"
What should come back is a short list, each name with the field it's based on, and a note about anyone it couldn't judge because they didn't say. Someone who answered "weekends: yes" and gave a current CPR expiry sits at the top. Someone who didn't mention transport gets a "not stated" next to it, not a strike against her.
Then: "Compare the top three." Rate, availability, location, experience, screening answers and any stored analysis, lined up side by side. It doesn't pick a winner for you. It lays it out so you can.
That's two questions and a couple of minutes, from the couch. You still ring them Monday morning. You still do the registry check. But you walk in knowing who to call first.
Questions that don't work so well
"Who should I hire?" It'll give you an answer, but it's the wrong question. You get a ranking built on whatever the applications happened to say, and the caregiver who wrote two lines from her phone ends up at the bottom for reasons that have nothing to do with care.
Questions about things nobody was asked. If your form doesn't ask about Sunday availability, "who can do Sundays?" turns into a guess at what people meant, and a decent assistant will tell you it can't really answer.
And questions that are really about the future. "Who's going to stay?" "Who'll actually turn up?" Nothing in an application answers those. That's what the interview and the reference call are for, and these are the questions I'd ask.
A few honest cautions
The applicant's details go to whichever AI you're chatting with. That's how it works. If your privacy notice to applicants doesn't cover it, fix that first, and if it sits wrong with you, don't turn it on.
Chat assistants can still get things wrong when they summarize. Treat the answer as a pointer back into First Shift, not a replacement for opening the application before you make the call.
And it's only as good as what came in. An application form that never asks about weekends can't tell you who does weekends. The questions on your form matter more than the AI does, which is why the job description template and the interview scorecard spell out what to ask.
The rest of the product is on how it works, and it's from $49 a month.
