⭐️ Top Pick of the Week

🔎 An insightful analysis of online experimentation practices, addressing the ' #ExperimentationGap ' within organizations.

Read the Towards Data Science article by Davis Treybig here 👉 towardsdatascience.com/the-experiment…

AI software engineering Copilot tools have taken the world by storm over the past few years - Github Copilot is nearing $1B in annual revenue! And yet, it feels to us like we are just at the very beginning of the transformation that AI will have on software engineering.

Augment is a truly full stack company, innovating from models and serving infrastructure all the way up through code representation, information retrieval, and the application layer.

As such, we are super excited to announce our co-lead investment into Augment's Series B (augmentcode.com)

Augment is building the next generation of AI assisted software tooling, starting with a next generation Copilot but quickly moving far beyond. Their initial product is fantastic and loved by the teams using it - sign up to try it out here: augmentcode.com/waitlist

See their TechCrunch announcement here: techcrunch.com/2024/04/24/eri…

Excited to collaborate with Sutter Hill Ventures Index Ventures Lightspeed and Meritech Capital on this

The company is led by one of the best teams we've ever met - Igor Ostrovsky is the former Chief Architect of Pure Storage, Scott is the former CEO of Pure Storage, and Guy Gur-Ari was one of the AI leads at Google behind much of pioneering work in language modeling.

Software is the most labor constrained market in the world and we think one of the largest companies ever can be created in this space. We're so excited to see what this amazing team that blends AI research, systems engineer, and product discipline can do.

The company also has a practical crawl, walk, run philosophy - not jumping straight to fully autonomous software agents but rather slowly increasing the scope and complexity of their product in a way that engineers can trust and understand.

Why do you need to pause to wait to see a code completion due to inference latency constraints? Why can these products only reference the 10-20 most recent files you have looked at at inference time? Why are half of their suggestions non-compilable?

There is so much opportunity to not only improve on the core Copilot form factor, but to also go dramatically beyond that into workflows that look more like a peer programmer helping you, all the way to a semi-autonomous junior engineer.