This week’s question comes from someone building their first multi-agent setup: “If one AI agent does the research and another does the writing, can the second one check the first one’s work?”
Short answer: yes, but only if it checks against something other than the first agent’s words. This week mine caught 4 wrong claims in 3 briefs. Here’s how.
The short answer: yes, if it checks against a source
In my fleet, researcher agents write briefs and writer agents turn those briefs into posts and videos. The writer’s instructions don’t say “follow the brief”. They say “verify the brief”. Every fact gets checked against the original source at write time, before anything goes live.
The checker isn’t a smarter model. It runs on the same model as the researcher. The difference is that it has a different job and it goes and reads the source.
What did my agents catch this week?
Three briefs, three days, four claims that didn’t hold up:
- Wednesday, the blog brief. It said not one search result mentioned that Claude Chat and Cowork had merged. When the writer agent re-checked, two pages already covered it, one of them from 13 October. The published post doesn’t say “nobody is talking about this”.
- Thursday, the video brief, claim one. It told the story of the Philae comet lander with the angle “plan your degraded mode before launch”. The writer checked Wikipedia, which doesn’t say that. It says the team chose what to run first on 14 November, as the battery ran down. The script used the version the source supports.
- Thursday, the video brief, claim two. Its receipt was “every JonOps skill has an Error Handling block”. The writer counted the skill files, and only 12 of 39 do. So the video quoted one real rule instead of a claim about all of them.
- Friday, the blog brief. It said my email agent works 2 inboxes. The writer read the skill file, and the agent works one. The post that went live overnight says one.
0 of the 4 reached a reader. None of them was a wild hallucination, either. Each was a plausible summary that went one step past its evidence, and that’s the kind of mistake a human skimming a brief would wave through too.
⚡ GET THE AI EDGE
Weekly AI tips that actually save you time and money. No fluff, no hype — just what works.
Why doesn’t “please double-check this” work?
Because a reviewer that only reads the brief has nothing to compare it to. Ask an agent “is this right?” and it checks whether the brief sounds consistent with itself, and it usually does. All four of these claims sounded fine.
Each catch happened because the checker went somewhere else: a live search, the Wikipedia page, a count of the actual files, the agent’s own instructions. It’s the same rule as treating an empty result as a claim, not a fact. A claim gets tested against the thing it describes, not against the document it came in.
I covered the bigger picture in the hallucination myth. The short version is that trust comes from the checking you build around the agent, not from the agent itself.
Do this today
- Make every brief carry its source. A link, a file path or a record ID next to each fact. If a claim has no source attached, the next agent treats it as unverified.
- Tell the checker to look, not just read. Re-fetch the page, re-count the rows, open the file. “Review this” gets agreement. “Verify each fact against its source” gets corrections.
- Give it permission to drop a claim. If the source doesn’t back it up, the claim goes, even when it was the brief’s best line. Two of my four catches were the brief’s main hook.
- Log what changed. My agents note every claim they corrected in their run notes. Those notes show me the upstream researcher keeps over-reaching, and they’re where this post came from.
The takeaway: one AI agent can fact-check another, but only against the source. Without it, you’ve just added a second agent that agrees with the first.
Want to see how the handoffs are wired? Here’s how my agents pass work to each other without a platform. If you’d like a verify step built into your own setup, book a strategy session.
📥 FREE: THE AI PLAYBOOK
The exact tools and workflows I use to run a one-person agency. 25 years of marketing experience distilled into an actionable guide. Yours free.
AI & Digital Marketing Strategist
With over 25 years in digital marketing, e-commerce, sales, and international business, Jon Jones is a seasoned entrepreneur dedicated to empowering small businesses through innovative AI solutions. His journey began in investment banking on Wall Street, guiding companies through capital acquisition and complex markets, and has expanded to include expertise in web development, brand design, product development, and courses and training.
Fluent in Mandarin and with a rich foundation in both finance and tech, Jon launched JonJones.ai to provide accessible, transformative tools for small businesses to streamline operations, generate leads, and build sustainable growth. His mastery of AI, automation, and digital strategy has enabled numerous businesses to bridge the gap between traditional practices and modern technology, ensuring they thrive in today’s digital landscape.
