The screen your team works in
A review queue, a tracker, a dashboard or a form built for the job, or the chat and inbox your team already opens. It is where people see what the agent did and sign off on it.
Proven agents, yours to keep
Switchboard Agents
Proven AI agents for the recurring work that eats your team’s week. We go through our catalog with you, set up the right ones on your systems and your rules, and stay until they run.
- Running in 2 to 4 weeks
- Your systems, any model
- A person approves what matters
- Runs on your accounts, yours to keep
Your week, back
The same jobs come back every week. Hand them over.
At a company of 100 to 1,000 people, a lot of the week goes to work that follows the same steps every time: matching, chasing, copying, reporting. Each job is too small to justify a software project and too frequent to ignore, so it lands on your best people. It is also exactly the work an agent can take.
One ops team, one week
The agents take the steps. Your people keep the approvals, the welcome and the judgment, and get the rest of the week back.
More than a chat box
Most agents are not a chatbot. They come with a screen.
Some jobs suit a chat window. Most suit the place the work already happens: a review queue, a tracker, a dashboard, a record, an approval in your inbox. So every agent we set up comes in three layers, and we build all three.
An exploded drawing of three stacked layers. On top, the interface: an app window with a queue of work on the left, each row carrying its own status, and a detail card on the right where a finished result has landed. In the middle, the logic layer: six steps wired in two branches that both run forward into a final step. Three are marked with a star and run on AI, two are rules, and one carries a person and stays a human decision. Underneath, the data layer: one wide record of fields that every link from the layers above lands on, ringed by eight of the tools the team already works in, two set into each edge of the layer, drawn as an inbox, a chat window, a calendar, a set of files, a spreadsheet, an invoice, a contact record and a database, each wired into the record. A job drops from the interface into the logic layer, down to the data, and comes back up as a finished result.
AI does the steps. People make the calls.
The steps, your rules and the AI that reads, matches and drafts. Anything outside the rules stops and waits for the person you named.
The systems you already run
Accounting, CRM, HR, ticketing, email and files, read and written by the agent within the access its role allows. Nothing is copied into a new tool to make it work.
The catalog
Start from an agent that has already done the job.
Every agent in the catalog comes from work we have built and run for clients. Here are two for each team. Pick one to see the screen your team would work in, and watch the agent run inside it, up to the moment a person steps in.
- Invoice matchingMatches each invoice to its PO and receipt, and posts the clean ones as drafts.
- Collections follow-upWorks the aging report every morning and chases what is overdue, in your tone.
- Inbound lead briefResearches every new lead, writes it to the CRM, and routes it to the right rep.
- Quote checkChecks every quote against your price book and discount rules before a prospect sees it.
- Request intakeReads internal requests, checks them against your approval rules, and gets them done.
- The Friday reportPulls the week’s numbers from every system and drafts the update before anyone asks.
- New hire onboardingTurns a signed offer into accounts, payroll, equipment and a day-one plan.
- Policy and leave questionsAnswers the questions HR gets every day from your own handbook, with the source linked.
- Ticket triage and refundsReads each ticket and its attachments, resolves what is within your limits, and escalates the rest.
- Conversation reviewReviews every call, chat and email against your standards, instead of a sample.
- Control evidenceCollects the evidence for every control, checks it is complete, and names what is missing.
- Auditor requestsAnswers an auditor’s request list straight from the record, with the source attached.
Also in the catalogAnd more, depending on how your business runs.
- Three-way reconciliationFinance
- Close checklistFinance
- CRM clean-up after callsSales
- Commission calculationsSales
- Cross-system handoffsOperations
- Internal questionsOperations
- OffboardingHR
- Headcount reconciliationHR
- Call transfer with contextSupport
- Promise check on repliesSupport
- Filing calendarCompliance
- Reportability checkCompliance
- One for a job only you do
The full catalog is bigger than this page. We go through it with you on a call: which agents fit your business, where to start, and how you’d want each one set up.
Walk through the catalog with us
What you get
Everything it takes to run, not just the agent.
An engagement ends with six things you can point at, whichever agent you start from.
- 01
The agent, set up for you
Picked from the catalog with you, then set to your process: your triggers, your naming, your exceptions and your approval limits.
- 02
Connected to your systems
Wired into the tools the job touches, with its own login and only the access its role needs.
- 03
Your choice of model
A frontier model, an open-weight one on your own servers, or an account you already pay for.
- 04
Approvals and a full log
The people you name sign off where you say, and every run records what it read, did and who approved it.
- 05
A scorecard
One number, today’s baseline and a target, agreed in week one and checked at go-live and after.
- 06
Your team, ready for it
A working session with the people who sit next to it, and a one-page runbook: what it does, who owns it, how to pause it.
Set up for you
Proven where it counts. Yours everywhere else.
The steps that make an agent reliable stay the same from client to client. Everything that makes it yours gets set in the first two weeks: which systems it reads, which rules it follows, who signs off, and what it writes back.
Your model, your call
Bring your own model, or pick one.
An agent runs on whichever model suits the job and your rules: a frontier model from OpenAI, Anthropic or Google, or an open-weight model like DeepSeek or GLM that can run on servers you control. Already paying for one? We use that account. Switch later, and the agent keeps its steps, its rules and its approvals.
Built for oversight
You decide what it can touch, and what waits for a person.
Anyone on your team should be able to answer three questions about an agent at any time: what can it see, what can it do without asking, and what did it do yesterday.
It sees what its role allows.
Each agent gets its own login with the access a person in that seat would have. The invoice agent reads accounting and purchase orders, and payroll stays out of reach.
You set where it stops.
Pick the actions that need a person and the thresholds that trigger them. Below the line it gets on with the work. Above it, it asks the approver you named and waits.
Every run leaves a record.
What it read, what it changed, who approved it and when. The person who owns the workflow can read it, and so can your auditors.
One person on your team owns each agent and can pause it from one screen. We build inside your security and approval rules from day one, the same way we build everything else.
How it runs
One call, then running in two to four weeks.
Every engagement follows the same path. A job on one system with a few rules finishes at the short end; more systems and more rules take the full four weeks.
- 0Before we start
Walk the catalog
One call. We go through the catalog against your business, pick where to start, and sketch how you’d want it set up.
- 1Week 1
Agree the number
Walk through the job with the people who do it, and write down the outcome and today’s baseline.
- 2Week 2
Set up and connect
Your rules go in, your systems get connected, and permissions and approvers are set.
- 3Week 3
Run alongside
It works the real queue next to your team. People check every output, and we tune what it gets wrong.
- 4Week 4
Live and measured
It takes the work on, approvals stay where you set them, and we check it against the week-one number.
From your side
- The people who do the job today. A few hours in week one and week three.
- One owner who signs off the scorecard and can pause the agent.
- Someone who can grant access to the systems involved.
From ours
- A lead who runs it from the catalog call to go-live.
- The engineers who connect your systems and set the agent up.
- The same people all the way through, and after launch if you keep us on.
Ways to start
Start with one agent. Add the next when it pays.
Every size gets the same six things. The difference is how many jobs, and how many teams.
One agent
2–4 weeks- Best when
- One job is eating a team’s week, and it is already in the catalog.
- Scope
- One agent, set up, connected, run alongside your team and live against its scorecard.
A team’s set
First live in 2–4 weeks- Best when
- One department runs on repeat work: finance at month end, HR every time someone joins.
- Scope
- Two to four agents for one team, sharing its connections and going live one after another.
Across the business
Ongoing- Best when
- Several teams have jobs in the catalog and you want a steady pace, not one big project.
- Scope
- A running list of agents across teams, the next one picked from how the last ones are doing.
After go-live
We stay on, or step back. Your call.
Keep us close and we look after the agents: we tune them when a system or a rule changes, check the scorecard with you each month, and bring the next one from the catalog. Or your team runs them and we step back. You own the setup either way.
Where the catalog comes from
Every agent was a client build first.
We kept building the same jobs inside bigger projects for mid-market teams. The ones that proved themselves became the catalog.
- People operations1 dayFrom offer accepted to first login.In the catalog asNew hire onboardingRead the case study
- Maritime operations25 hrs/wkReclaimed from email triage.In the catalog asRequest intakeRead the case study
- Agency operations35%Faster invoice cycle.In the catalog asInvoice matchingRead the case study
We’re builders who advise. This is that, packaged.
How we compare
Plenty of agents for sale. Few that fit.
A product built for everyone, a project for a team you don’t have, or a plan with nobody to run it. We set up the agent and stay until it works.
Switchboard Agents
Off-the-shelf agents
Build it yourselves
Consulting firms
What you get
An agent running on your systems, set up to your rules.
A product you configure yourselves.
A project for people you would have to hire.
A strategy deck.
Fit
Your process, approvals and exceptions, written in.
Their defaults and a settings page.
As deep as you have time for.
Outside-in.
Speed
Running in 2 to 4 weeks.
Fast to buy, slow to fit.
Months, if it ships.
Months of discovery.
Model
Your choice, including your own.
Theirs.
Whatever you can wire up.
A recommendation.
Controls
Permissions, approvals and a log from day one.
What the vendor ships.
Whatever you remember to build.
In the recommendations.
After launch
We stay on, or your team runs it.
A support queue.
You are the support team.
The engagement ends.
Questions
What people ask before they start.
What is a Switchboard agent?
Software that does one recurring job from start to finish. It starts on a trigger, reads the systems the job needs, does the steps, and hands anything outside your rules to a named person. Each one starts from a version in our catalog that we have already built and run, then gets set up for your business.
How is this different from an AI Jumpstart?
A Jumpstart is for when you are still working out where AI belongs, and it ends with a map and a working prototype. Agents are for when you already know the job, and they end with the agent running on your systems. Plenty of clients do a Jumpstart first and pick their agents from what it finds.
Do we need engineers on staff?
No. We do the setup, the connections and the testing. From your side we need the people who do the work today for a few hours a week, and someone who can grant access to the systems involved.
Which systems can an agent connect to?
Most of what a mid-market company runs: accounting, CRM, HR and payroll, ticketing, shared drives, email and spreadsheets. If a system has no way in for software, we tell you in week one, before anything gets built.
Which AI model does an agent run on?
Whichever suits the job and your rules. That can be a frontier model from OpenAI, Anthropic or Google, an open-weight model such as DeepSeek or GLM running on servers you control, or an account your company already pays for. You can switch later without changing the agent’s steps, rules or approvals.
What happens when the agent gets something wrong?
Anything outside the rules you set goes to a person instead of going through. In week three the agent runs next to your team and people check every output, so the rules get tuned on your real work before it takes anything on by itself.
Who owns the agent after launch?
You do. It runs on your accounts under the permissions you set, and one person on your team owns it and can pause it. We can stay on to look after it and add the next one, or hand it over.
What does it cost to run once it is live?
Hosting and the AI it uses, both scaled to your volume. We estimate them in week one from your real numbers, so you know the running cost before you decide to go live.
Can you build an agent that is not on the list?
Yes. If the job is close to one we have built, we start from that. If it is new, it becomes a custom build through the AI Deployment Studio.
Next step
Got a job eating your team’s week?
Tell us what it is. On one call we’ll go through the catalog with your business in mind, show you the agents that fit, and sketch how you’d want the first one set up. You leave with a starting point whether or not we work together.