Agent Ops Log #0: What It Cost to Set Up an AI Agent Team
ByteandRise is run by a team of AI agents with a person approving what ships. This is the first issue of the Agent Ops Log, a monthly report on what that team produced, what it cost and what broke.
ByteandRise is run by a team of AI agents with a person approving what ships. This is the first issue of the Agent Ops Log, a monthly report on what that team produced, what it cost and what broke. Issue #0 covers the setup: getting ten agents, a website and a publishing pipeline working together, and the first full content run.
The team we set up
The agent team runs on Claude Platform as managed agents. One coordinator, the CEO agent, plans the work, hands tasks to specialists and reads their reports. Each specialist has one job and the smallest model that does it well:
| Agent | Job | Model tier |
|---|---|---|
| CEO | Plans, delegates, reports to a person | Large |
| Strategist | Positioning, messaging, proof rules | Large |
| Tech Lead | Site code, API, migrations | Large |
| Editor | Quality gate before anything reaches a person | Large |
| SEO Lead | Topics and briefs | Mid |
| Writer | Drafts | Mid |
| Designer | Implements approved designs | Mid |
| QA | Tests changes | Mid |
| Social | LinkedIn text for the company page | Mid |
| Analyst | Search Console and analytics numbers | Small |
Agents share two memory stores: a company brain (positioning, voice, proof points, banned claims) and an ops store (tasks, briefs, reports, handoffs). Every task ends with a written report, so the next run starts from what the last one learned instead of from scratch.
Nothing goes live on its own. The Editor scores every package, and a person approves it in the admin panel before it is scheduled. You can see the whole flow on how we run ByteandRise.
What it cost in October
All numbers are list prices from the platform's usage data, as of mid-October.
| Where the money went | Runs | Cost |
|---|---|---|
| Setup sessions (site, positioning, design, QA) | 4 | $44.31 |
| First content cycle | 1 | $4.90 |
| Total | 5 | $49.21 |
By agent, the Tech Lead was the largest line at $29.21: it did the heaviest coding work during setup and runs on a large model. The CEO came second at $7.43, then QA ($3.46), Designer ($2.86), Strategist ($2.11), SEO Lead ($1.84), Editor ($1.67) and Writer ($0.48).
One decision kept this low. One-off build work, such as the site rebuild and the admin panel, runs in a coding assistant session on a flat subscription, not on the agent platform. The agent team is reserved for recurring work: content, reviews and reporting. Paying per token for a one-time rebuild would have cost far more than it saved.
The first content run
The first scheduled content cycle ran on October 13. The CEO picked an approved topic, the SEO Lead wrote a brief, the Writer drafted, and the Editor checked every claim and the on-page SEO. Fifteen minutes of active work produced a 2,006-word guide, What Is an AI Agent Harness?, which the Editor passed at 90/100 and a person approved for publishing.
That run cost $4.90:
- CEO: $1.48
- SEO Lead: $1.25
- Writer: $0.48
- Editor: $1.67
The surprise is the shape, not the total. Writing the post was the cheapest step. Coordinating, briefing and checking cost about nine times as much as drafting. That is the price of the quality gate, and we think it is worth paying: a cheap draft that a person has to fix is not cheap.
What broke
The agents couldn't reach our own API. We moved the site to new hosting, but the agents' sandbox kept resolving the domain to the old server from a cached DNS record. The first pilot run stopped at its first API call. We updated the old DNS zone as well as the new one, and the next run went through.
A style rule leaked into the admin panel. A class name from the public site's background animation also existed in the admin panel, and turned every form field into an absolutely positioned block. Topic details overlapped into an unreadable stack. We renamed the class and scoped it.
The cost page said $0.00. The admin panel had a costs page from day one, but nothing was feeding it: the sync from the agent platform was never wired. We built the sync and backfilled October, which is where the numbers above come from.
The first run nearly hit its cap. Each content run has a $5 hard limit. Spending $4.90 means one more web search or one more revision would have stopped the run halfway.
What we'll change next month
- Bring a content run under $4. The SEO Lead's brief cost more than the Writer's draft. We'll tighten the brief template and check whether the coordinator needs a large model for routine runs.
- Track cost per published post. With content running Monday, Wednesday and Friday, cost per post will be the number to watch.
- Watch what the gate catches. The Editor's holds and reasons will be in the next issue, so you can see where agents still get things wrong.
If you are planning an agent of your own, the AI agent cost calculator uses the same building blocks: tokens per call, calls per task, review time and overhead. To get the next Agent Ops Log by email, subscribe to the monthly email.
FAQ
How much does it cost to run an AI agent team?
For us in October: $49.21 at list prices, of which $44.31 was one-off setup and $4.90 was the first recurring content run. Recurring work is what matters month to month, and we cap every run.
Why does a content run cost more than the draft itself?
Most of the cost is coordination, briefing and checking: in our first run the Writer cost $0.48 and the CEO, SEO Lead and Editor together cost $4.40. That is the price of a quality gate before a person sees the draft.
Does anything publish without a person approving it?
No. The Editor agent scores every package and a person approves it in the admin panel before it is scheduled.