Work smarter
Flynn finds, matches and drafts for your consultants. You decide how it sounds, how far it may go and what waits for a person, and you can try a change first or undo one after. Need a new helper? Describe it, and switch it on once you are happy.
The problem
You would like the help, but not an AI that says or does things in your agency’s name without anyone checking.
How Vextron fixes it
Your agency chooses Flynn’s tone, the phrases it must never use, how bold it may be on a judgement call and which actions wait for a person. Nothing it drafts sends itself, and money is out of its reach.
What you gain
You get the help without handing over the keys.
The problem
Changing how an AI behaves feels like a gamble. You only find out it was wrong once a client has seen it.
How Vextron fixes it
A change can be run as a practice before it is let loose. If an agent starts doing worse than your agency’s normal, you are told, and its changes can be previewed and rolled back.
What you gain
A wrong setting is caught by you, early, rather than by a client.
The problem
The small, repetitive jobs never justify a developer, so they land on your best people instead.
How Vextron fixes it
An admin describes the job and Flynn sets up the agent: when it runs, what it does and what it may touch, explained as it goes. It runs once a person approves it, and it can be a step in your workflows.
What you gain
The chores your agency has put up with for years finally get a helper.
The problem
Finding the right person means building filter after filter, and you still miss people already in your database.
How Vextron fixes it
Flynn searches candidates from a plain-English question and narrows the results as the recruiter pushes back, bringing back people who have gone quiet.
What you gain
More placements come from candidates you already know.
The problem
A score knocks someone out, and nobody can say why.
How Vextron fixes it
Candidates are scored against a job with the reasoning shown, so a recruiter can disagree. No one is rejected by an AI score without a person reviewing it.
What you gain
Your consultants stay in charge of every decision.
will it go rogue?
Everything Flynn and its agents may do is set in one control room, by your agency. Step through it, or let it play.
AI Command CenterHarbour & Co · Settings
Choose the tone, note how you like things put, and list the phrases Flynn must never use. Every draft follows them, before anyone reads it.
Clients hear your agency, not a chatbot, whoever is busy that day.
Decide how bold agents may be on a judgement call, and whether they only suggest or act once a person approves. Acting alone isn’t on the menu.
The AI never goes further than your most careful manager would.
Run a change as a practice first. See exactly what would be different, with nothing touched and nobody contacted, then switch it on or leave it.
You find out a setting is wrong here, not from a client.
When an agent wants to do something, the full action waits in a queue: what, for whom, and why. Approve it or turn it down.
Every action has a person’s yes on it before it happens.
If an agent’s results slip away from your agency’s normal, you hear about it. Look at what it changed, and roll it back.
A bad week for an agent stays a small problem.
Tone
Writing style notes
“Short sentences, no jargon, always sign off with first name only.”
Banned phrases · tap to ban or allow
To Paul G. · Logistics firm
Draft · waits for youHi Paul,
I have two finance managers for your role: Priya D. and Tom H. Priya is a rockstarproven lead at month-end.
Could we touch basespeak on Thursday? I can line up interviews ASAPthis week.
Jo
Hi Paul,
Good news: two finance managers I think you’ll like, Priya D. and Tom H. Priya’s a rockstarnatural at month-end.
Fancy a quick touch basechat on Thursday? Happy to set up interviews ASAPthis week.
Jo
Paul,
Two finance managers for you: Priya D. and Tom H. Priya is a rockstarspecialist at month-end.
Touch baseCall Thursday? Interviews ASAPthis week.
Jo
Hi Paul, I hope the week’s been kind to you.
I’ve met two finance managers I’d love you to meet, Priya D. and Tom H. Priya is a rockstarreal expert at month-end.
Could we touch basetalk on Thursday? I can arrange interviews ASAPthis week.
Jo
Dear Paul,
I am pleased to introduce two candidates for your Finance Manager role: Priya D. and Tom H. Priya is a rockstarspecialist in month-end close.
Might we touch basespeak on Thursday? I can arrange interviews ASAPthis week.
Jo
Rewritten the moment a setting changes.
Risk appetite
A judgement call
Maya O. for the Payroll manager job: a year short on the experience asked for.
Left out. Only clear fits are suggested.
Suggested, marked borderline, with the reasons.
Suggested alongside the clear fits, reasons shown.
What agents may do
Money is out of reach. No agent or automated step can touch pay, bills or invoices.
Change waiting to go liveCandidate scorer · risk appetite Balanced Bold
The practice runs on this week’s Payroll manager job, Logistics firm. Nothing is changed and nobody is contacted.
Practising on the Payroll manager job…
Now · Balanced
With Bold
Practice only: nothing changed, nobody contacted.
Switched on: Bold, for the candidate scorer. One more borderline name on this job, reasons shown, and a person still decides.
Kept as it was: Balanced. Nothing went live.
Waiting for a person
Each action shown in full before anything happens
Candidate scorer
Move Priya D. to Interview
Finance Manager · Logistics firm
Why:Led month-end close for four years; open to a move
Approved by you
Declined. Nothing happened.
Flynn
Put Sam K. forward for Friday night
Picker packer · Packing site
Why:Has worked this site before; documents up to date
Approved by you
Declined. Nothing happened.
Job matcher
Add a task for Leon W.: call Alex R.
Business analyst · contract ends in four weeks
Why:Two open roles fit his skills and day rate
Approved by you
Declined. Nothing happened.
Your agents
Job matcher · success rate
Last 3 weeks
Job matcher has drifted below your normalSince Tuesday’s update. Look at what it changed.
Since TuesdayMatches re-ranked on 6 jobs · 11 suggestions added
Rolled backIts changes since Tuesday are undone.
describe it, Flynn builds it, you approve it
An admin writes what they want in plain English. Flynn turns it into an agent you can read line by line, and a person switches it on. Pick a request to try.
Agent builderAI Command Center · for admins
1You describe it
Here’s your timesheet chaser, part by part. It won’t run until someone approves it.
Here’s a next-role finder. Every part is below, in plain English. Nothing runs until it’s approved.
Done: an agent that checks your own database first. Have a look before you switch it on.
2Flynn builds it
Timesheet chaser
New agentNext-role finder
New agentDatabase first
New agent3You approve it
Flynn is putting it together. Nothing is running.
Timesheet chaser is ready.Not running yet.
Approved by youOn. Every Friday at 3pm its reminders land in your approval queue.
Or drop it into a workflow
Flynn is putting it together. Nothing is running.
Next-role finder is ready.Not running yet.
Approved by youOn. Its suggestions land on the consultant’s task list.
Or drop it into a workflow
Flynn is putting it together. Nothing is running.
Database first is ready.Not running yet.
Approved by youOn. Its top five land on each new job for the consultant to review.
Or drop it into a workflow
Ask Flynn
Ask the way you would ask a colleague. Flynn searches your whole database, ranks the people who fit with its reasons shown, and drafts the outreach. Nothing goes out until you send it.
Try asking
Pick a question and Flynn will answer it here.
I found 14 people in your database. These three fit best, with my reasons:
Push back and I’ll narrow it, for example to people free in the next month.
Five people fit the Packing site nights and haven’t been booked for over six weeks. The closest three:
Want me to draft a message to them? Nothing goes out until you send it.
Six contractors in your database are free next month and fit the developer contract. Here is a draft for the closest, written in your agency’s tone. They all wait here until you send them.
To Jordan T.
Subject A six-month developer contract from next month
Hi Jordan,
I’m working on a six-month developer contract with a software firm in Manchester, on a day rate, starting next month. Your dates line up well. Have you got ten minutes for a quick chat this week?
Thanks, Jo
+ 5 more drafts, for the other contractors who fit
Placeholder: the video and real early-adopter stories are on their way.
See Flynn in two minutes
A short walk through an agency setting Flynn up and putting it to work: the rules, a new agent, and a consultant’s morning with it.
Tone, limits and approvals set in the AI Command CenterA new agent described in plain English and approvedMatches with their reasons, and drafts that wait to be sent
correct → learn
The candidate scorer and job matcher learn from what your consultants actually shortlist, accept and turn down. You can see what each agent has picked up from your corrections, and none of it ever reaches another agency.
search → shortlist
Flynn searches, matches and drafts personalised outreach, and guides a recruiter from search to placement. It appears in Flynn Hub, in its own panel and on records.
ask → explained
Beyond building agents, admins get plain-English help designing forms and creating custom metrics, with each choice explained as it goes. And when anyone needs help using Vextron, the support assistant is upfront that it is an AI.
Seven AI agents work inside Vextron: Flynn, a candidate scorer, a job matcher, an agent builder, a forms architect, a metric architect and a support assistant. The scorer and the matcher cannot delete anything, each declares exactly which records it may read, create or update, and nothing either of them touches is financial.
In the AI Command Center an agency can see which agents are on, what they have done, how each is performing, and what each has learned from the agency’s own corrections. Drift alerting raises an alert when an agent’s results move away from the agency’s own normal.
Placeholder: longer write-up for search engines (about 400 to 600 words) to come.
No. Every agent either suggests, or acts once a person approves; acting entirely on its own is not offered. Outreach Flynn drafts waits for someone to send it, nobody is rejected by an AI score without a person reviewing it, and no agent or automated step can touch pay, bills or invoices.
Try it first. An agent’s run can be simulated before it is let loose, and an agent’s changes can be previewed and rolled back. If an agent’s success rate or speed moves away from your agency’s normal, the AI Command Center raises an alert.
Yes. An admin describes what they want in plain English, and the agent builder sets out what the agent will do, when it runs and what it may touch, explaining the technical parts. It runs once a person approves it, and it can be called as a step in a workflow.
Yes. Your agency sets Flynn’s tone, adds notes on how you like things written and lists any phrases it must never use. Every draft follows them.
No. An agent learns only from its own agency. Nothing an agent learns in one agency reaches another.
Yes. The support assistant has a non-human name and says it is an AI assistant in its first line.
see it for yourself
A demo built around the desks you run. Vextron is in early access.