Krishna Katragadda

Professional bio at

[Krishna Katragadda @krishckat]

@gokulr Stated intent rarely matches revealed intent as the void is filled with cognitive biases. Misread intent leads to wrong product. Instead shouldn't it be objective and goal driven as we anyway have OKR for alignment. We don't have any metrics for intent alignment.

[Krishna Katragadda @krishckat]

@KanikaBK Try adding infranodus in Obsidian. It'll find new connections that you might be missing. It's like expanding the pie.

[Krishna Katragadda @krishckat]

@Srini_Pa @vkhosla No question about human flourishing and climbing up the Maslow's hierarchy of needs using AI and other next gen technologies that expand the pie.

Its about abundance that could derail.

[Krishna Katragadda @krishckat]

"You can never offload your learning" is the whole game. But the loop still assumes the decision left a trace to absorb. In the highest-stakes settings it doesn't, the bed that wasn't called, the consult deferred, the trauma bay held open. The most expensive decisions a firm makes emit no signal at the moment they're made. Compounding starts there, or it starts late.

[Krishna Katragadda @krishckat]

@paulg Does this apply to LLMs? Did we anticipate it in the first place or purely accidental? 🤔

[Krishna Katragadda @krishckat]

@anshublog @amir It's operating income over revenue, which is operating profit margin or simply operating margin is -122% (operating loss). So for every $1 revenue their operating income is $-1.22

On a $10B revenue, their operating income (loss) here is $-12.2B(EBIT)

[Krishna Katragadda @krishckat]

We know what experience is and assume that experience can be tapped in by humans with the click of a button relying on AI.

What is the experience? Have we fully comprehended what experience really is or confined ourselves to a narrow definition that we are comfortable with. This could be a philosophical article on its own.

On the biological side, experience is not just pattern matching or judgement, but also the level of information compression that comes from experiences.

I recently learned from Emergency physicians, where some had 30 years while others had less than 10 years of experience treating patients in a chaotic environment.

Their level of comfort in handling 10+ patients simultaneously is something that they can only achieve through the years of experience. Experience in this case is not just judgment, but treating more patients simultaneously by compressing the amount of information, putting aside the emotional rollercoaster they experience.

Their experiences are more than judgment. They have rewired their brains through neuroplasticity and strengthened their emotional regulation, which is an asset in making better decisions.

Also, when we talk about experience, we only focus on what's codes in 0s and 1s. Can everything be coded in 0s and 1s other than assigning a numerical value to pain on a scale of 1 to 10. The pain on a scale of 1 to 10 might be 7 for two people but their experiences, level of tolerance are different.

An experienced physician reads the unlabeled signal: body posture, affect, what isn't said, whether the number matches the person holding it. That signal was never converted to data. It exists in the encounter, not in any record. AI cannot retrieve what was never stored.

What does experience mean to you?

Is experience your identity or uniqueness that we all strive to be in this human race?

Experience = Experience + X.
Do we know what that X is?

[Jaya Gupta @JayaGup10]

https://t.co/yiGW8jKKXk

[Krishna Katragadda @krishckat]

Enterprise work and decisions are not just contextual. They are stateful, temporal(time bound), and bounded by information availability at the point in time.

Decision makers or agents optimize within limits of state, context and time, often referred to as bounded rationality, where limits themselves move and the decision's validity decays as state evolves. That's why some of the decisions look subpar or bad when looked at with complete information or during an after action review.

[Krishna Katragadda @krishckat]

NASA doesn't let flight directors get interrupted without protocol.

Emergency physicians get interrupted 12.5 times per hour. 84 times per shift or almost every 5 minutes. And 18.5% of interrupted tasks are never resumed.

No error message. No alert. The cognitive thread just vanishes.

In software engineering, one interruption costs 23 minutes of recovery time. An emergency physician managing 15 patients simultaneously doesn't lose 23 minutes. They lose the entire cognitive thread for a patient's care trajectory.

Nobody catches it.

Here's what CMIOs need to understand: the bottleneck in ED patient safety isn't knowledge. It's working memory.

A BMJ Quality & Safety study found that below-average sleep increased ED clinical error rates by more than 15x. Working memory capacity was the only mitigating factor. A 10-point increase on a standardized working memory test produced a 19% reduction in procedural error rates.

The CDS infrastructure you built assumes physicians fail because they don't know things. That's the wrong diagnosis. They fail because they can't hold things. Normal capacity is 7 plus or minus 2 items in working memory but they manage around 15 patients. 100+ micro-decisions per hour. The math doesn't work.

NASA solved this in the 1960s with distributed situational awareness. The flight director holds the meta-graph. Specialized controllers hold the subgraphs. No single human holds the full mission state.

Your EHR doesn't enable this architecture. It's a documentation system, not a situational awareness system. It records events after they happen. It doesn't track what's pending, what's at risk of being dropped, or what the physician should attend to next.

The highest-value informatics investment isn't better CDS alerts, which are overridden 90% of the time anyway due to notification fatigue.

It's the cognitive infrastructure layer between the physician and the operational complexity of the department.

AI documentation tools already showed a 21.2% burnout reduction in 84 days at Mass General Brigham. That's one narrow intervention. The full cognitive offloading opportunity is orders of magnitude larger.

You've spent years optimizing data capture. The actual bottleneck is the human processing it.

Here's my question for CMIOs: If you could instrument one invisible metric in your ED right now, what would it be? Pending action aging? Task dropout rate by hour of shift? Cognitive load proxy by patient volume and acuity mix?

The answer determines where the informatics strategy should actually aim.

If you're already thinking about how to architect this layer, DM me. I've been mapping what this looks like across different EHR ecosystems and would rather build on your real constraints than abstract ones.

#EmergencyDepartment #CognitiveOverload #ContextGraphs #PatientSafety

[Krishna Katragadda @krishckat]

@alexiskold Does founder conviction show up in metrics that VCs look for?

[Krishna Katragadda @krishckat]

An airplane descends through 800 feet of solid cloud. The pilot can't see the runway.

Yet, the aircraft lands safely. Precisely on the centerline in the touchdown zone.

Because the pilot has an Instrument Landing System(ILS).

The localizer keeps them aligned laterally. The glide slope holds them on a descent path. Marker beacons confirm progress. At decision height (200 feet), they see the runway and land, or they go around.

The ILS doesn't fly the plane. The pilot flies the plane. The ILS provides the cognitive infrastructure that makes safe landing possible when human senses alone are insufficient.

Now consider the emergency physicians.

They're managing 15 patients simultaneously. Each patient is an approach in progress, a trajectory from arrival to disposition. The "landing" is the disposition decision: discharge, admit, or go around or observation.

Every approach is being flown in zero visibility, not physical but cognitive.

12.5 interruptions per hour. 84 per shift. 18.5% of interrupted tasks never resumed.

That's an 18.5% rate of losing the localizer signal mid-approach and never reacquiring it. In aviation, that grounds the fleet. In emergency medicine, it's Tuesday.

The Emergency Department(ED) has no ILS equivalent:
No localizer (Is this patient on the right clinical trajectory?)
No glide slope (Is this patient progressing toward disposition at the right pace?)
No marker beacons (Which patients are stalled between clinical milestones?)
No decision height (When must I commit to disposition or escalate?).

The physician holds all of this for all 15 patients in 7 plus or minus 2 working memory slots as knowledge and context graphs. Managing 60-90 concurrent cognitive items through compression alone.

Then every 8 hours, the knowledge graph is destroyed.

Verbal handoffs retain 0 to 26% of information. 74% of hypoxia events go unreported at shift change. The incoming physician inherits the equivalent of a cockpit mid-approach, zero visibility, no approach plates, no ILS frequency programmed.

Emergency medicine has a 52.2% burnout rate (AMA 2024), Highest of any specialty.

The burnout isn't from clinical work. It's from flying 15 simultaneous instrument approaches in zero visibility with no instruments. Every shift and for an entire career.

Diagnosis isn't the bottleneck, but coordination is. The physician gets the diagnosis right. The glide slope deviation that harms the patient is the delayed antibiotic, the missed consult callback, the disposition deferred because cognitive load hit the ceiling.

Aviation built the ILS 70 years ago. Emergency medicine still lands patients on human cognitive infrastructure alone.

Question for medical directors: What's your department's version of "losing the localizer"? The moment a patient's trajectory falls out of the physician's cognitive field and doesn't come back?

Your pattern recognition reveals which ILS component would prevent the most harm.

[Krishna Katragadda @krishckat]

@paulg Portmanteau words - A word formed by merging the sounds and meanings of two different words have a higher chance of finding AI domains...

[Krishna Katragadda @krishckat]

@lydlyd @levie Anyone can build the context infrastructure but the real moat is the compounded collection of tacit knowledge that doesn't get captured in the current data systems.

[Krishna Katragadda @krishckat]

@KirkMarple Frameworks like Organizational Ontology and Organizational Structure Ontology can be leveraged in conjunction with Domain Ontology, utilizing knowledge graphs and hypergraphs across the temporal dimension to facilitate context-aware decision-making.

[Krishna Katragadda @krishckat]

@clairevo @OpenAI @chatprd Better than Opus 4.1?

[Krishna Katragadda @krishckat]

@AladdinMhir @lovable Dashboard

[Krishna Katragadda @krishckat]

@AladdinMhir @boltdotnew Hack

[Krishna Katragadda @krishckat]

@damienghader @lovable 🚀

[Krishna Katragadda @krishckat]

This post explores different aspects of human capabilities that will have an impact on decision-making at companies using the metaphor of a boat navigating choppy waters.

Steering the Corporate Boat: A 360° Look at Decision Intelligence - The Human side
https://t.co/sRNkcgsNRz