Sanjay Siddhanti

I wrote a post about deploying Python web apps on Lambda. It includes practical guidance on how to migrate a DB, serve static files, store secrets, and use Docker. Captures my experience learning serverless architectures and initially being disoriented. medium.com/@sanjay.siddhanti…

As every engineer has re-learned their craft over the last year, companies should be questioning assumptions about how teams operate. We've been able to staff teams that are far leaner now. Empirically, this has worked better than giving coding agents to a traditional team and expecting proportional improvements. If 1-2 engineers feel ownership for an entirely new product, they are forced to shed their identities as specialists, and use the best tools available to fill gaps. In this model we've seen designers shipping frontends, frontend engineers shipping backends, or even ML researchers building their entire labeling app from scratch. Meanwhile in a traditional team, human tendency is to revert to your identity as a specialist, and ask other people for help. Also there are more toes to step on and less permission to move fast.

With all the excitement about Fable, here's how it affects healthcare companies: TLDR Fable 5 is HIPAA compliant for *API usage* but not within Claude Code. For API usage, you have to disable Zero Day Retention and possibly upgrade to a new type of BAA. They now require 30 day retention for Fable usage to flag safety issues, which is awkward at best for healthcare companies. They do commit to not training models on your data, which is good. In general a lot of the best Claude features - including Cowork, remote control, and now Fable - are not fully HIPAA compliant. This is a huge bummer and I hope they make quick progress on this.

As someone who hates making slides, Claude has been a game changer. I've tried a few tools but this is the first one that's saving me a lot of time. What's working for me recently: - Claude Chat to research my Slack and Google Docs and iterate on the talking points - Claude Powerpoint plugin to build the slides. I find it works *much* better when I give it a skeleton deck with the right fonts / colors already baked in - Upload back to Google slides when done Uploading back and forth to Google Slides is annoying, but it works so much better than the Gemini integration in Google Slides right now, so it's worth the friction

Spent the day at the Pragmatic Summit in SF hosted by @GergelyOrosz . We're seeing a fundamental change in what it means to be an engineer and how we build teams. Here are a few of my takeaways: 1. Generalists > specialists Everyone has read Karpathy's post on hiring for agency. In product development, that's a full stack generalist who can use AI tools effectively. PMs and designers should be building prototypes and shipping code, not writing PRDs and mocks. People who love the craft of writing code by hand need to adapt rapidly and refocus on the increased business impact they can now deliver. 2. Smaller, flatter teams The traditional "two pizza team" concept needs to shrink. We are seeing more 1-3 person teams that can move faster due to less overhead. Teams should push more ownership to the engineers; having a manager in the loop on every decision will slow down velocity. 3. Single player -> multi player Most of us have had our "Claude code moment" and have achieved immense productivity improvements for solo or greenfield work. Now we need to scale that across multiplayer, multi-repo projects. The frontier labs have a head start which we can learn from. First, teams need to expose the agents to critical context that often lives elsewhere (design docs, product requirements, etc). Second, teams should invest in shared repositories of skills, tools, and workflows that every developer can benefit from. Find the early adopters and codify their workflow into something that everyone else can use. 4. Execs should do IC work again More than ever, leaders and execs need to ship code again. We all use our experience as builders (often many years ago) to gut check technical approach, output, and velocity. Those mental models need to be rebuilt in today's world. Fortunately, it's never been easier to get back into writing code and everybody should be doing it.

Ashby feature request: expose an MCP server so customers can talk to an LLM about pipeline strength, candidate feedback, speed to offer, etc. I'm hiring for 20+ roles and this capability could replace multiple meetings at my company. Seems like a great opportunity for building something "your CTO will use" 🙂 Your support team told me that API keys cannot be scoped to specific roles, so I can't let an EM pull data without giving them access to all roles at the company. @benjaminencz is this on your roadmap? Happy to provide feedback if so!

So awesome to see the next generation of engineers and data scientists in @akasahealth swag!

It was so fun to go back to my high school and visit the AP CS classes! @kickrg was a huge positive influence in my life and his classes led me to start thinking about studying CS and becoming a software engineer

Check out my blog post about how we run virtual Hackathons at @akasahealth. I discuss why employers should prioritize Hackathons, how to plan them, and how to hack collaboratively during the pandemic. akasa.com/blog/akasahacks-ho…

So excited about our fundraise at AKASA! We're looking for great software engineers to join the team. We have tons of interesting problems re: RPA, data engineering, and computer vision. This team ships products quickly and you can make an immediate impact in RCM. DMs are open.

Adding to this - some of the best engineers I’ve worked with didn’t go to top schools or don’t have CS degrees