Caleb Gawne

Learning, building.

Co-founder of Kye AI. Former @RoseRocketApp, @veevasys.

[Caleb Gawne @cdgawne]

@levie Two other points:
- US labs can extend their lead as they're the only ones with access to true frontier models.
- It manages labs' costs for them. They don't have to serve the largest models and can spend more resources distilling models before GA.

[Caleb Gawne @cdgawne]

@paulg I expect reading without an LLM summary becomes equally rare. The only unfiltered reading worth the time will be from human authors.

[Caleb Gawne @cdgawne]

@mikeknoop In easily verifiable domains?

[Caleb Gawne @cdgawne]

@fordsmith Our largest to date is CultivateBHE, a healthcare provider with ~1000 providers in 40 locations with a legacy EMR, spreadsheets, emails etc.

Although the tech can scale beyond that, we're focused on the mid-market. They're the most underserved at the moment.

[Caleb Gawne @cdgawne]

The first step is wrangling all activity and data across business apps without building hundreds of integrations. Kye AI 's been mapping out processes, quantifying bottlenecks and deploying AI agents built on this data for 2 years in healthcare, logistics and financial services.

The final mile (and the hardest part IMO) is the data that can't be captured because:
-It can't be extracted (eg. physically impossible)
-What employees choose to obfuscate for various reasons
-What the company chooses not to extract for compliance, legal reasons

[Tom Blomfield @t_blom]

Imagine replacing 90% of your employees with a team of geniuses who have no idea how your company operates.

Total chaos. Nothing works.

That’s what AI feels like today.

The missing piece is extracting all the domain knowledge from people’s heads and providing that as structured context to the models.

[Caleb Gawne @cdgawne]

@jedpolglase @t_blom There's:
-What can't be extracted
-What employees choose to obfuscate, or outright mislead as they realize job security rests in this information
-What the company chooses not to extract for compliance, legal reasons

[Caleb Gawne @cdgawne]

@jordanschochet @t_blom What are these domains?

While every industry has easily-verifiable tasks, I can't think of any large industries where you have a highly-reliable verifier for entire roles, never mind companies.

[Caleb Gawne @cdgawne]

@paulg @t_blom In purely digital services companies yes. If you're an AI-native company with a physical presence then there can still be domain knowledge hidden in people's heads.

Employees will likely increase information hordeing to increase job security as well.

[Caleb Gawne @cdgawne]

@t_blom The first step is at least wrangling all activity and data across business apps. Kye AI did this a couple years ago (even before MCP).

Extracting the WHY of decisions and processes is the true final mile though. I haven't seen any scalable solutions for this yet.

[Caleb Gawne @cdgawne]

@stevesi @levie To get to 'a company as a unique AI model...' there's still a substantial amount of tribal knowledge that's not digitally captured at WMT and most organizations. For example, WHY did the dispatcher choose to send this parcel with this carrier.

[Caleb Gawne @cdgawne]

@levie You don't need every single system to be integrated to capture the full business context across spreadsheets, legacy apps etc. Kye AI solved this a year+ ago in healthcare and logistics for any business app or website, even if their's no MCP.

[Caleb Gawne @cdgawne]

@Austen Kye AI does AI audits using weeks of screen captures across teams of 100+.

Redaction and app/website filters are a must though.