In partnership with
Perfect vitals, sweaty patient
On one chest pain call, my patient's vitals looked perfect even though he was sweating heavily. The monitor said stable, the patient said otherwise, and the 12-lead we ran on every chest pain call showed a heart attack in progress. IBM recently asked employees whether they feel safe questioning AI output when they think it's wrong, and 82% said yes, while their executives guessed 59%. My short answer is yes, mostly, because people are already skeptical of AI and we're told to distrust it. But 43% of employees say the blame for an AI error lands on them. This one is about the difference between feeling safe to question an output and feeling safe once it goes wrong, and the conversation to have after the next AI mistake that reaches real work.
Last issue, a Claude agent noticed a missed payment and held off on saying so, because the agent that owed it had always paid before. This one is about the people on your team when an AI output that has always been right gets one wrong.
Stable on the monitor
I had a chest pain call where the patient was extremely sweaty and his vitals looked perfect. Heavy sweating is one of the classic signs of a heart attack, and a lot of the time the monitor can show stable while the patient is saying otherwise. Patients can be compensating well until they're not, and then they crash fast. You also can hedge bets and just assume when the electrodes have difficulty sticking to their skin and we’re just sitting there, it’s not normal. We typically start giving medication while we’re struggling with the sweat, to at least start doing something helpful. Yet, we still run a 12-lead, which shows far more of the heart's electrical activity than the 4-lead does, because we run one on every chest pain patient. Lo and behold, he was having a STEMI, the kind of heart attack where a coronary artery is blocked. However, this isn’t always the case and still requires due diligence in questioning because some circumstances can be explained - such as heavy drug use.
Nobody on that call needed courage to doubt the monitor. The 12-lead was protocol, so the second look happened whether the first one looked fine or not.
As the lead, I carried ultimate responsibility for that patient. Technically everyone providing care is responsible, but the lead is the one held to it. When you're right, it's an unsung-hero kind of moment. Sometimes a teammate pats you on the back, sometimes the hospital does, and rarely does the patient know to thank you.
It was always safe to say something looked wrong. There's no room for error on a 911 call, so raising a disagreement quickly mattered more than being sure first, and if the disagreement turned out to be wrong, that was fine. Both sides learned something, and the concern got considered either way. That was baked into our culture. I can't promise it's baked into every EMS agency or fire department, but it was in ours.
That safety had an edge, and it sat at the hospital doors. In the field we were all equals, and that was important. At the hospital, we couldn't really question the receiving physician in the moment. We could raise it afterward as a learning tool, never during the handoff.
That's also why Power Tripping went the way it did. The captain who pulled rank on me ran a different organization with a different culture, and his probationary firefighters, taking instruction from the person running the scene, had none of the cover I had to speak up.
Two kinds of safe
By feeling safe to question, I mean nobody judges you for it and the team is already interactive and deliberate. Blame tends to live in teams that don't work as one unit and operate in silos.
In IBM's 2026 CHRO study, 82% of employees said they feel safe questioning or ignoring AI output when they think it's wrong. Executives, asked to estimate the same thing, pictured a far more hesitant workforce at 59%. That doesn't match the idea that employees are afraid to push back on AI. In my experience people are skeptical of AI already, and we're told to distrust it.
But 43% of employees say that when AI goes wrong, the blame lands on them. Feeling safe to question an output and feeling safe once the output proves wrong are two separate things.
The need also changes by team and by moment, the way it did at the hospital doors. Some AI decisions are field calls, where anyone should be able to say "this looks wrong" while the work is happening. Others are closer to the handoff with the receiving physician, where the question waits for the review afterward. Sort your team's AI decisions into those two kinds before the next mistake.
In the same study, 71% of CHROs name digital and AI fluency as their top development priority. IBM also found that 72% of organizations make limited or no use of AI inside HR itself, and only 13% of executives rate HR's AI literacy as highly capable. HR is still learning AI, so most of this work falls to the teams.
The monitor stayed on for the whole call. When the 12-lead disagreed with it, we acted on the 12-lead.
The Vitals
The read. IBM Institute for Business Value, Designing the Thinking Organization, 2026 CHRO Study, September 2026.
Who they asked. 1,500 CHROs and senior executives in 21 geographies, and separately 8,800 full-time employees in 28 countries, surveyed April to June 2026.
The finding. 82% of employees said they feel safe questioning or ignoring AI output when they think it is wrong. Executives estimated 59% of their employees would say so (Figure 4, p. 36).
The supporting line. 43% of employees report that when AI goes wrong, the blame lands on them (p. 18).
How to read it. Both numbers are self-report from separate samples, so read the 23 points as a difference in perception between two groups. IBM describes its relationships as associations and makes no causal claim (p. 49).
Source: IBM Institute for Business Value, Designing the Thinking Organization: 5 focus areas for CHROs as AI reshapes work, 2026 CHRO Study.
The Handoff
The one conversation this week: after the next AI mistake that reaches real work, hold a round table the way we did after hard calls. Ask what the AI showed and what the team saw, and keep the conversation on the work, with blame left out of the room.
From the build
SteadMark observes the work already happening in a team's chat channels and turns it into a read a leader can act on, so they can see whether each team is living the change they planned. Reads are team level, message text is deleted once it's classified, managers confirm or correct what it finds, and people make every decision. A read is meant to be questioned, like any AI output.
This is my seventh newsletter. Hit reply with anything and I’ll be happy to chat.
With love,
Jim Sullivan
Stop feeding the feed.
You’ve put in the work to build an audience. But when the only way to reach your fans is through a social feed determined by algorithms, the platform decides who sees what you make next.
beehiiv gives you the tools to reach your audience directly. On their all-in-one platform, you can turn followers into subscribers, grow your newsletter, and connect it all to your community or podcast. Stop chasing virality on platforms you can’t control. Build a stronger, more sustainable business on beehiiv.