AI Studio

AI Studio is where you decide how much of your ad operations you want to hand over — and to whom. It holds four building blocks: Adlyse AI, the assistant that already knows your accounts; Agents, the colleagues you hire and brief yourself; Workflows, fixed pipelines that run the same way every time; and Schedules, the clock that starts any of it without you.

The mental model#

The four blocks answer four different questions. Reaching for the wrong one is the most common way teams end up with automation they don’t trust.

Read it as a ladder of how much you’re fixing in advance. Adlyse AI fixes nothing — every conversation is fresh. A workflow fixes everything — the same nodes in the same order. Agents sit in between: fixed identity and tools, free judgment.

Which one should I reach for?#

Start with Adlyse AI#

Adlyse AI is already wired to your accounts, your brand profile, your files and your skills. It has the safest configuration in the product and it dispatches to platform specialists for you. For anything one-off — a question, an investigation, a change you want to make now — it is the right and cheapest answer.

Use it when: you’re exploring, diagnosing, or making a change you’ll approve on the spot.

Spin up a custom agent when the job has an owner#

A custom agent is worth the setup when a job recurs with a point of view attached to it — a standing brief, a narrower tool set, a specific model, a voice. You are hiring a colleague, not writing a script: you tell it what it’s responsible for and what it may touch, and it works out the steps each time.

Use it when:

• The job repeats and you keep re-typing the same context (“you’re our ecommerce specialist, our target ROAS is 3.2, never touch brand campaigns”).
• You want it to run unattended on a schedule, so there’s nobody to answer a follow-up question.
• You want it narrower than Adlyse AI on purpose — a compliance reviewer that can read everything and change nothing.
• You want it to hand work to sub-agents it owns.

Don’t build an agent just to ask a question once. And don’t build one to lock down a sequence — that’s a workflow.

Build a workflow when the steps must not vary#

A workflow is a graph you draw: fetch this, branch on that, call an agent here, email the result there. Nothing about the order is left to a model’s judgment. That’s the whole point — it’s what you want for anything auditable, anything with a hard branch, and anything that touches a system outside Adlyse.

Use it when: the sequence is the requirement, or the job needs steps that aren’t AI at all — an API call, a code step, a conditional, a human approval gate in the middle.

Add a schedule when it should happen without you#

A schedule is a clock pointed at an agent or a workflow. It adds no intelligence of its own — it just means you stop being the trigger.

What automation can and can’t do on its own#

This is the single most important rule in AI Studio, and it’s a safety property, not a limitation:

An unattended run can never change your ad platforms directly. A scheduled agent that spots a wasteful campaign cannot pause it. What it can do is write a proposal card into your decision queue — and into Slack, if you’ve connected it — with the numbers and the reasoning attached, waiting for a human to approve.

The reason is simple: every real platform change goes through an approval preview, and on a background run there is nobody there to answer it. So instead of failing silently or acting without consent, the run leaves you a decision.

A worked example#

You want budget pacing reviewed every Monday.

1. Try it in Adlyse AI first. Ask “review budget pacing across all campaigns and tell me what to reallocate.” If the answer is good, you’ve just proven the job is possible and you know what a good output looks like.
2. Promote it to an agent. Create a Budget Analyst, paste your target ROAS and your don’t-touch list into its instructions, give it the ad-platform integration plus the Suggestions capability, and nothing else.
3. Point a schedule at it. Weekly, Monday 9am, with the same instruction you tested in step 1.
4. Work the queue. Monday morning the proposals are in Home with dollar impact attached; you approve the ones you like.

Notice that step 1 is not optional. The fastest way to build automation you trust is to run the job by hand once and steal the prompt.