ai:onboarding
AI onboarding for new hires: start people on your workflows, not a blank tab.
By Loan Laux · September 21, 2026 · 9 min read
AI onboarding for new hires means folding your agency's AI workflows into the same first weeks that already cover the PM tool and the brand deck: a sanctioned seat provisioned before they arrive, the usage policy walked through on day one, and their first real deliverable routed through the workflow defaults with a buddy reviewing. Done properly, a new hire's first proposal, ticket batch, or pull request comes out of your system. Skipped, they default to whatever they did at their last job, usually a personal ChatGPT tab and habits you'll spend a year untraining.
This assumes there is something to onboard people into. If AI at your agency is still individual experimentation, fix that first (start with getting your existing team to use AI); onboarding transmits a system, it doesn't create one.
Why onboarding beats every other adoption tactic.
Getting existing employees to change how they work is a slog, and we've written about what it takes: defaults, visible leadership, honest answers about jobs. All of it fights the same two enemies, habit and fear. A senior strategist has ten years of muscle memory and a reputation as the expert of the old way; asking them to be a beginner again is a real cost.
A new hire has neither enemy. They have no workflow at your agency to defend, no standing to lose by asking basic questions, and they arrive expecting to be told how things are done. Onboarding is the one window where 'this is how we work' lands as orientation instead of disruption. Miss it and you've converted your easiest case into another retrofit project.
It also compounds. A few years of turnover and growth later, much of the team either joined through your onboarding or improvised around it. Every hire onboarded onto the system strengthens the defaults; every hire left to improvise dilutes them.
What improvised AI onboarding looks like.
Most agencies don't decide to skip AI onboarding; they just haven't updated the checklist since 2022. The new PM gets the PM tool login, the brand guidelines, and a Slack welcome, then quietly opens the personal ChatGPT account from their last job. Three things follow: client context in an unsanctioned account, output that sounds like generic AI instead of your agency, and a first month spent re-deriving prompts the team wrote down a year ago.
The gap between that and a designed first month is not subtle:
| First 30 days | Improvised | Designed |
|---|---|---|
| Tools | Personal account, whatever they used before | Sanctioned seat with SSO, live before day one |
| Rules | Guessed from watching colleagues | Policy walked through in 30 minutes, in writing |
| Prompts | Re-invented from scratch | Shared library and skills, toured in week one |
| First deliverable | Drafted from a blank prompt | Routed through the workflow default, buddy-reviewed |
| Day 30 | Private habits you can't see or fix | Visible work shipped through the system |
AI onboarding for new hires: the first 30 days.
01:
Before they arrive: provision the seat
A paid seat on the sanctioned assistant, SSO connected, integrations installed on their machine, and the usage policy in the welcome pack next to the contract. If the sanctioned tool is ready before a personal one is open, you've won the default.
02:
Day one: walk the policy, don't send it
Thirty minutes with a person, not a PDF: which tools are sanctioned and paid for, what client data can and cannot go in, what clients are told about how the agency uses AI. New hires are the most careful people in the building; ambiguity reads as prohibition, and prohibition sends them back to the private tab.
03:
Week one: tour the artifacts
The prompt library, the skills that encode your processes, the SOPs with the AI step written in, the knowledge base. Not a demo of what AI can do; a tour of what this agency has already built, and where to put anything they create.
04:
Week one: shadow the workflow, live
They sit with someone in their role doing the real thing: a proposal drafted from the generator, tickets pulled from a call transcript, a PR with generated tests under review. Watching a colleague work the default teaches more than any workshop, because it settles what's normal here.
05:
Week two: first deliverable through the default
Their first real task goes through the AI-shaped workflow, reviewed by a buddy before it goes anywhere. The point is not speed yet; it's that their very first artifact at your agency was made the way you make things.
06:
Weeks two to four: buddy and office hours
Pair them with the role's strongest AI user for stuck moments, and put them in the same weekly office hour the rest of the team uses. Habits form or die in this stretch, and it costs almost nothing to hold.
07:
Day 30: check deliverables, not enthusiasm
Look at the work artifacts: shipped through the defaults or around them? Prompts pulled from the library, anything added back? If the artifacts don't show the system, extend the buddy period rather than waiting for a review to find out.
What each role's first week should cover.
:
Project managers
The meeting-to-ticket flow, status reports assembled from the PM tool, and first-draft client emails in the account's voice, each practiced on a live account they'll actually run.
:
Designers
Brief expansion and variant exploration on a current project, plus the handoff checks. The message to land: AI speeds the work around the craft; the craft is why they were hired.
:
Developers
The agency's coding-assistant setup and review standards: context discipline, test-first prompting, and what generated code must pass before it reaches a PR. Their first assisted PR should happen inside week two, reviewed by whoever owns the standard.
:
Account leads
Reading AI-drafted work critically before it reaches a client, and the exact answer this agency gives when a client asks how AI is used on their account.
The artifacts that do the teaching.
A designed onboarding is mostly a tour of things that already exist. Four artifacts carry the weight:
- The usage policy. One page, in plain language: sanctioned tools, data rules, client disclosure. If yours doesn't exist yet, write it before the next hire starts.
- The prompt library and skills. The difference between 'here's ChatGPT' and 'here's how we draft scopes'. Claude Skills are the sturdier version: process encoded once, invoked by anyone, including someone in week one.
- SOPs with the AI step written in. A new hire follows the SOP by definition. If the SOP says the proposal starts from the generator, the default installs itself. AI-era SOPs are the cheapest onboarding tool you can build.
- The knowledge base. Where past work, decisions, and client context live, and what your retrieval workflows draw on. A usable knowledge base is what makes a new hire's AI output sound like your agency instead of the internet.
Mistakes that undo it.
- Scheduling the AI part for month two. By then the private habits are set and you're back to running an adoption project on one person. The window is the first two weeks, while everything is new anyway.
- Assuming the young hire 'knows AI'. Fluent personal use is not client-safe use. They may prompt well; they don't know your data rules, your voice, or your review standards, and confidence makes that gap more dangerous, not less.
- Making it a lecture. A capability tour changes nothing in onboarding for the same reason it changes nothing in training: watching is not working. Every session should end with the new hire having driven a real task.
- Onboarding onto tools instead of workflows. A list of logins is not a system. The unit of onboarding is 'how a proposal happens here', not 'here's our ChatGPT plan'.
- Leaving the buddy unnamed. 'Ask anyone' means ask no one, especially for someone new who is rationing their questions. One named person, for one month.
Do it yourself, or bring us in.
Everything above is buildable internally: add the AI items to the existing onboarding checklist, name the buddies, and spend a few days assembling the artifacts if they don't exist yet. The artifacts are the honest hard part: agencies that already have a policy, a prompt library, and SOPs with the AI step written in can wire this up in a week; agencies that don't discover that their onboarding problem is really a systems problem.
That systems work is what our forward-deployed AI engineers do: build the workflows and artifacts with your team, on live client work, and leave them in your accounts, at which point onboarding the next hire is a tour instead of a project.
Frequently asked questions.
How long should AI onboarding for new hires take?
No extra calendar time: it folds into the onboarding month you already run. The AI-specific parts are thirty minutes of policy on day one, a few hours of artifact tour and shadowing in week one, and a named buddy through day 30. Sequencing matters more than volume; the sanctioned setup has to arrive before the private habits do.
What if the new hire is better at AI than we are?
It happens, and it's a gift with a catch. Harvest what they know: have them add their best prompts to the library and demo one technique at the weekly ritual. But they still learn your defaults, your data rules, and your voice, because fluent personal use is not the same as client-safe agency use.
Should freelancers and contractors get AI onboarding too?
A lighter version, yes, and arguably more urgently: contractors are the most likely to put client context into a personal account and the least likely to have read your policy. The minimum: the policy walked through, a sanctioned way to work, and the defaults for whatever they're delivering.
What if we don't have AI workflows to onboard people into yet?
Then that's the project, and hiring is a good forcing function for it. Pick the one workflow per role with the clearest win, write the SOP with the AI step in it, and let the next hire be the first person onboarded onto it. If you're unsure which workflows those are, an AI workflow audit exists to rank them.
Who should own AI onboarding?
Whoever owns onboarding now, with role champions supplying the content: the ops lead runs the checklist, the strongest AI user in each role does the shadowing and buddy duty. Creating a separate 'AI onboarding' owner just builds a silo around something that should be ordinary.