ScheduleMon 09:00
Agency referee: 3 findings this week
Two campaigns are stuck in learning; one losing campaign is still scaled.
- Woke on
- Mondays, 09:00
- Read
- Meta Ads · Google Ads
- Next
- Open all three findings
Build agents that report back on their own.
Agents are Adgent's custom checks that wake on your terms and report back on their own. Add your own agents to the team working toward your marketing goals. Pick the sources, say what wakes each one and what to look for. Every run returns with the finding, evidence and next move in Outputs.
Three ways to wake oneOutputsThis week
Choose a run to see why it woke, what it read and what happens next.
One live account, 30 Sep 2026: a request to referee the agency's work became the proposed weekly agent.
Three ways to wake one.
Say it in a sentence: when to run, what to read, what to look for. The agent waits for its moment, then writes what it found, why it matters and the next move.
On a schedule
The check you'd ask for every Monday.
In a live meeting, one request to referee the agency's work became a proposed weekly agent: Mondays 09:00, into Outputs. Client reports and feed checks run the same way.
Agency referee
Mondays, 09:00
Reads Meta Ads · Google Ads
Mon 09:00 · 3 findings this week
On a condition
Quiet until the number moves.
Say it plainly: CPA 20% over its two-week average. It is built to wait until that holds, then write what moved and what it would change.
CPA watch
CPA 20% over its 2-week average
Reads Meta Ads · Google Ads · GA4
Thu 07:40 · 24% above, proposes a lower budget
On an event
When something new shows up.
A campaign is created; the agent checks its setup against the rest of the account and writes down what differs.
New campaign check
When a campaign is created
Reads Meta Ads · Google Ads
Tue 14:32 · “Autumn sale” optimises for link clicks
One sheet holds the whole agent.
Describe it in a sentence in chat, or fill the sheet yourself: name, sources, what wakes it, the instruction and where the output goes. It runs as an AI agent or a fixed rule.
The same sheet keeps the trigger, evidence sources, instruction and output format together, so the team can inspect or change the workflow without rebuilding it.
New agentDraft
Name
CPA watch
Sources
Meta AdsGoogle AdsGA4Search ConsoleMerchant Center
Wakes on
ConditionEventSchedule
Condition
CPA 20% or more above its 14-day average, for 3 checks
Instruction
“Find what drove the rise and tell me what you'd change.”
Runs as
AI agentFixed rule
Sends to
Outputs · PDF, CSV, Excel or HTML
May propose
Budget changesPausing an ad set
Owner review Account moves leave the agent as decision cards.
A finding becomes an account move.
The decision card keeps the evidence, current state and proposed state together. The completed move lands in the Change Ledger with its result.
From CPA watchMeta
Prospecting · daily budget
Current$40.00 / day
Proposed$36.00 / day
Owner review
07:52
Youfrom CPA watch
$40.00$36.00
Revert with Agent
The inverse starts from the ledger's recorded before-state.
Ten specialists come built in, one lever each.
They turn findings into ranked, priced decisions; your agents add checks on top.
- Budget AllocatorBudget split by habit, not result
- Account ArchitectDuplicated, fragmented structure
- Measurement IntegrityTracking you can't trust
- Audience-Overlap DetectorAd sets bidding against each other
- Placement & Network OptimizerWasteful placements, devices, geos
- CPC-Spike AutopsyWhat drove a CPC spike
- Creative-Quality DoctorIncomplete or weak creative
- Creative-Fatigue ScoutAds wearing out
- Delivery & Burn WatchSpend with no return, or no delivery
- Budget-Pacing CopBudgets running out or overspending
Questions about agents.
What happens after I describe an agent?
Adgent turns the sentence into a schedule or condition, selects the sources and sends each run to Outputs. Open any run to see the evidence and next move.
How do I build one?
Describe it in a sentence in chat, or fill in the settings yourself: name, sources, what wakes it, the instruction and where the output goes.
Won't it become another stream of alerts?
It is built to speak when its condition changes, then cool down. Runs collect in Outputs, where you mark each one seen or dismiss it.
What can an agent read?
Your Meta and Google Ads accounts, plus GA4, Search Console, Merchant Center and the Drive files you pick. What each source adds.
Tell us what you want Adgent to watch.
The agent team reads your accounts toward your goals, brings back what it finds and keeps the recurring work moving in Outputs.
- 1Connect Meta and Google AdsAdd GA4 or Search Console when useful.
- 2Adgent reads the accountWith GA4 and Search Console if you connect them.
- 3You get the decisionsRanked by what each fix is worth.