Latest / MBA Training AI / Fyxer: how an AI email agent earns user trust

Transcript

Fyxer: how an AI email agent earns user trust · MBA Training AI

0:00
Host: Welcome to the Leaders Insights Podcast.
0:07
Host: Today's episode, how Fixer built an AI executive assistant people actually trust.
0:13
Host: Everyone wants an AI agent running their inbox.
0:16
Host: Almost nobody wants to actually let one.
0:19
Host: Why is that gap so wide?
0:20
Expert: Because automation and delegation are different animals.
0:24
Expert: Automating means the machine does a task and you check the result.
0:28
Expert: Automating means you stop checking.
0:30
Expert: That second step is where every agent company either wins or dies.
0:34
Expert: And most die.
0:36
Host: Fixer built an AI executive assistant for email and calendar.
0:39
Host: What did they get right that the inbox cleanup crowd didn't?
0:42
Expert: They understood the product wasn't email sorting.
0:45
Expert: It was trust.
0:46
Expert: A human assistant earns your confidence over weeks, learning that you never reply to recruiters
0:52
Expert: and always fast-track your biggest client.
0:55
Expert: Fixer engineered that same arc into software instead of assuming users would tolerate a
1:00
Expert: stranger touching their inbox on day one.
1:03
Host: That sounds like a nice slogan.
1:04
Host: Concretely, how do you engineer trust?
1:07
Expert: Three moving parts.
1:08
Expert: First, fine-tuning.
1:10
Expert: Retraining the underlying model on your specific writing so the drafts sound like
1:14
Expert: you, not like a chatbot.
1:16
Expert: Second, persistent memory.
1:18
Expert: The agent remembers context across sessions instead of forgetting you every morning.
1:23
Expert: Third, structured human feedback, meaning every edit you make to a draft becomes a training
1:29
Expert: signal, not just a one-off correction.
1:31
Host: The memory piece worries me.
1:33
Host: An assistant that remembers everything also misremembers things at scale.
1:38
Expert: Correct, and that's the trap.
1:40
Expert: Memory without correction is just a confidently wrong colleague.
1:43
Expert: The fix is keeping the human in the loop early and pulling them out gradually.
1:48
Expert: Fixer starts by drafting replies you approve, then earns wider latitude as its
1:53
Expert: accuracy climbs.
1:54
Expert: You're not handing over the keys, you're extending the leash.
1:58
Host: Let's talk numbers.
1:59
Host: What does adoption actually look like when trust is done properly?
2:03
Expert: OpenAI has cited internal figures suggesting agents can handle a large share of routine
2:08
Expert: knowledge work, but flag that hard.
2:11
Expert: They sell the models underneath these products so their framing is generous.
2:15
Expert: Cross-check against MIT Technology Review, which has been far more sober, pointing out
2:21
Expert: that most enterprise agent pilots in 2026 stall before full deployment, precisely because
2:27
Expert: of the trust handoff.
2:28
Host: So the vendors are selling a future.
2:31
Host: The independent press says isn't here yet.
2:33
Expert: Exactly.
2:34
Expert: And Google AI publishes similar optimism about autonomous agents.
2:38
Expert: Again, they're pushing their own platform.
2:41
Expert: So treat those adoption claims as marketing until an outside source confirms them.
2:47
Expert: O'Reilly Radar's developer surveys tell the truer story.
2:51
Expert: Engineers love building agents, but the ones shipping to real users obsess over guardrails,
2:57
Expert: not autonomy.
2:58
Host: Isn't there a contradiction here?
3:00
Host: You want the agent to act independently, but you're describing constant supervision.
3:05
Expert: The best operators already know it's not a contradiction, it's a sequence.
3:10
Expert: Trust is earned through repetition under observation.
3:13
Expert: Then autonomy is granted.
3:15
Expert: Same way you'd onboard a human hire.
3:17
Expert: Nobody gives a new employee signing authority in week one.
3:20
Expert: The mistake companies make is assuming AI skips that probation because it's fast.
3:25
Host: What happens to companies that skip probation anyway?
3:28
Expert: They ship an agent that sends one embarrassing email to a board member, and the whole organization
3:33
Expert: bans the tool.
3:34
Expert: I've watched it happen.
3:36
Expert: One bad autonomous action erases a thousand good ones because trust is asymmetric.
3:41
Expert: It builds slowly and collapses instantly.
3:44
Expert: That asymmetry is the entire design constraint.
3:47
Host: So the durable principle underneath Fixer isn't about email at all.
3:51
Expert: Not even slightly.
3:53
Expert: The principle is that any agent touching real stakes has to make its reasoning visible
3:57
Expert: and its errors cheap to catch.
4:00
Expert: Fixer's drafts-you-approve flow isn't a limitation, it's the product.
4:04
Expert: The moment you hide the reasoning, the user stops trusting the output.
4:08
Expert: And a distrusted agent is worse than no agent.
4:11
Expert: Now you're double-checking everything and paying for the privilege.
4:15
Host: Give me the one thing a listener building an agent should do Monday morning.
4:19
Expert: Design the feedback loop before the automation.
4:22
Expert: Ship your agent in draft mode, where every human correction retrains it, and only widen
4:27
Expert: its authority per task once its accuracy passes a threshold you set in advance.
4:32
Expert: Earn autonomy.
4:33
Expert: Don't assume it.
4:34
Host: That's the whole game.
4:36
Expert: And it's the part everyone's tempted to skip.
4:38
Expert: Earn the leash before you drop it.
4:40
Host: That'll do it.
4:41
Host: Sources for today's episode.
4:43
Host: Ours Technica AI, Google AI, Uvendor, AI Lab, MIT Technology Review, The Decoder, O'Reilly
4:51
Host: Radar, KD Nuggets.
4:53
Host: Done.
4:54
Host: Want to know where you actually stand?
4:56
Host: Take the AI self-assessment at MBA-training.com.