Three Rovo agent archetypes take the same twenty minutes to build in Agent Studio. Only one of them survives contact with the credit math. Every tutorial on the web — Atlassian’s own included — walks you through clicking Identity, Subagent, Tools, Knowledge, Activate. None of them tell you that the form doesn’t discriminate between an agent worth having and one that quietly eats your org’s entire monthly Rovo allowance by the second week. This page does that judgment call: three concrete agent archetypes, what each actually costs to run at realistic volume, and which one is worth building at all.
The number that actually decides this
Per Atlassian’s usage-allowance documentation, a Rovo Agent run costs a flat 10 credits, no matter what the agent does — reads, writes, tool calls, or a plain knowledge lookup answered in Chat. A Deep Research call costs 100. Plain search, summaries, and definitions are free, but the moment you’ve configured a Studio agent to answer a question, that answer is billed as an agent run, not as free search.
Set against that flat 10-credit cost, the monthly allowance is 25 credits on Standard, 70 on Premium, 150 on Enterprise, per user, pooled at the organization level. Do the division and you get the number that should anchor every “should we build this” conversation: about 2.5 agent runs per Standard user per month (7 on Premium, 15 on Enterprise) before an org is drawing on other people’s unused allowance. Rovo pricing and credits, explained covers the full allowance model; the piece missing from that page — and from every build tutorial — is what happens when you hold a specific agent idea up against that ceiling.
The honest version: build effort tells you almost nothing about whether an agent is worth it. All three archetypes below are equally trivial in Agent Studio — no Forge, no custom code, no more than a handful of built-in tools. What separates them is how many times a month a real team would actually call them, multiplied by 10 credits.
Archetype 1: the status/rollup agent — the one that pencils out
The job: “Summarize epic X across its 20 linked issues” or “what’s blocked in this sprint and why.” Two built-in tools cover it — JQL search to pull the linked work, optionally a Confluence read for context — well inside Studio’s tools list and nowhere near the 5-tool guidance ceiling.
The economics: this is naturally low-frequency, human-paced work. A PM checking on a handful of epics three times a week is roughly 12 runs a month — 120 credits, about a fifth of one Standard user’s full monthly allowance, for the highest-synthesis-value task of the three. Even five PMs doing this constantly stays well inside what a normal-sized org’s pool can absorb without anyone else giving anything up.
Verdict: build it. It’s the case Rovo’s pricing was actually designed for — broad, occasional, high-value calls — and the effort-to-build is identical to the other two archetypes, so there’s no reason not to.
Archetype 2: the triage/routing agent — pencils out only if you gate it
The job: a new ticket lands, the agent reads it and decides team, priority, or component, then updates the fields or reassigns it. Also cheap to build: JQL search plus an update-work-item action, maybe a Confluence knowledge scope for routing rules — two or three tools, still under Studio’s guidance. How to build a Rovo agent covers wiring this kind of agent to fire from a Jira Automation rule’s “Use Rovo agent” action so it runs on ticket creation rather than waiting for someone to chat with it.
The economics — this is the trap: unlike the rollup agent, a triage agent’s call volume isn’t set by how often a person feels like asking something. It’s set by ticket volume, and ticket volume doesn’t know or care what your credit pool is. Back-of-envelope, using Atlassian’s own published unit costs: a 20-person Standard-tier org fielding a modest 15 new requests a day that all route through the agent is 15 × 10 = 150 credits a day, roughly 3,300 credits a month — against a pool of about 500 (20 users × 25 credits). Triage alone would run at roughly 6-7x the org’s entire monthly Rovo allowance, before anyone opens Rovo Chat for anything else.
Verdict: worth building, but not as “the agent triages everything.” Have your Automation rule handle the obvious cases with plain field-matching — no credits, no agent — and only invoke Rovo for the genuinely ambiguous remainder it can’t resolve deterministically. That’s the version where a triage agent’s cost profile looks like the rollup agent’s rather than like a runaway meter.
Archetype 3: the knowledge/FAQ agent — the one that feels free and isn’t
The job: point an agent at a Confluence runbook or product-docs space and let people ask it questions instead of searching manually. This is, mechanically, the cheapest of the three to build — zero tools required, just instructions and a knowledge scope. If build effort were the filter, this would win by default.
The economics — this is why it isn’t the filter: question-asking is the highest-frequency behavior of anything on this page, by nature. It’s also the one place people most intuitively assume they’re “just looking something up” for free — but a configured Studio agent answering in Chat is billed as a 10-credit agent run every time, the same rate as writing to Jira. Twenty people asking three questions a day is 60 questions daily, about 1,320 a month, north of 13,000 credits — against that same 20-seat Standard org’s ~500-credit pool. That’s not a rounding error; it’s roughly 26x over, and it’s the archetype most likely to actually see that usage pattern once people realize it’s there.
Verdict: be the most skeptical here, not the least. The build being trivial is exactly why teams ship this one first and are surprised when the pool drains. If you build it, do it on Premium or Enterprise pool sizing, scope the knowledge tightly so answers stay short and don’t invite iterative back-and-forth, and watch org-level usage from week one — see what happens when the pool runs out before you’re the one finding out live.
Build effort was never the filter
All three archetypes fit comfortably inside what Agent Studio can do — none of them need the custom logic or external API calls that would push you into a Forge build. If you’re choosing between these three based on which is fastest to stand up, you’ll build the FAQ agent first every time, because it’s the easiest — and that’s precisely backwards from the economics. Worth checking too: none of the three run into Rovo’s documented failure modes on their own — those show up on complex JSM intake forms and long instruction sets, a different problem from the one this page covers.
Run your own team’s numbers — seats, plan tier, and a realistic monthly call count per archetype — through the Rovo credit cost estimator before you commit to any of these. It’s the same 10/10/100-credit math above, just with your actual headcount plugged in instead of a 20-person example.
Know which agent is worth building but don’t want to be the one debugging tool selection, knowledge scope, and automation-rule gating by trial and error? We take Rovo agents from spec to deployed, gated the way the credit math above argues for. Configure what you need and see the price and timeline instantly →