Igniting the torch of agent coded health tech to enable subject matter experts to build production grade software.
Joined February 2025
- Tweets11
- Following33
- Followers3
- Likes14
Nils Widal retweeted
I am similarly pumped to discuss org-wide AI-native adoption with @mihail_eric and @NWidal67643! Tragically, “give everyone Claude Code” is not an organizational transformation strategy, and the process of trial and error involves error. We have, however, learned a lot.
Pumped to discuss the challenges and learnings of org-wide AI-native adoption this Thursday.
Too many talks and articles discuss the nitty-gritty of tools, prompts, context engineering. Not enough talk about how coding agents will change the very DNA of engineering teams. That's what this panel is about:
- How adoption differs by scale of company
- Challenges of pushing AI-adoption org-wide
- The first 30 days of an AI-native push including important tooling and policy decisions
Looking forward to diving into all of these topics with Nils Widal, CTO of Cara, and @HendersGame, Senior Eng Manager at @honeycombio. @MavenHQ
30 min. Free. All signal, no noise. maven.com/p/c79c8d/challenge…. .
In case you are interested following, I will be building on the Claude Build Day today
github.com/carainc/cara-spar…
Heading to @claudeai build day for what was fable 5 and now is Opus 4.8
Kudos to @claudeai, @claudedevs, and @cerebral_valley for still holding the event and allowing us to build meaningful stuff!
One of the truest statements I have heard in a while: Vibe Coding does not democratize software engineering, it democratizes liability.
#vibecoding #agenticAI #codingagents
For everyone using Claude Code, sharing this open source repo for a light weight coding harness optimized for healthcare apps
github.com/carainc/reppit-he…
With this as a foundation and a great deal of preparation and planning
in a single run, our system:
• Executed 66 tickets across multiple repos
• Wrote 500+ tests
• Wrote 20K lines
• 19 parallel subagent batches
• Passed HIPAA, HITRUST, and SOC2 audits with zero deferred issues
The full writeup and paper is here widal.substack.com/p/we-ship…
#healthtech #codingagents #ai #healthcare
fwiw I wrote up the 4 hours autonomous run with @claudeai I did -- shipped a 66 ticket epic end to end to staging -- completely done by a coding agent harness and claude code, I am sharing what and how here: widal.substack.com/p/we-ship…
In clinical AI, "let the model decide" is not a safety architecture. It's a liability.
The LLM proposes. Deterministic code decides.
The policy engine is not an LLM tool. It is a function. π(evidence, risk, policy) → action.
The model can never override it.
Wish @doctronic would do it that way.
#ClinicalAI #HealthcareAI #PatientSafety #AIinHealthcare #HealthTech #MedAI #AIAgents #LLM #ClinicalSafety #DigitalHealth