The feature backlog has gone *poof*
Implications from the world waking up to agentic engineering this week.
Welcome to issue #64 of next big thing.
In a portfolio company board meeting last week with co-investor and friend Sarah Tavel at Benchmark, she said the following:
I was talking to another founder who said they suddenly have no feature backlog. The amount of output that an engineer can have has grown so much just in the last month that, all of a sudden, the feature backlog has gone poof.
What’s Sarah referring to?
There seems to be broad consensus in Silicon Valley that something really changed in the last few months. The coding models from Anthropic (Opus), OpenAI (Codex), Gemini, Cursor, Cognition (Devin), and more got really good. Not just at generating code, but at accomplishing coding tasks that require longer time periods and context windows. Andrej Karpathy posted about this on X on January 26th, describing his own experience with Claude Code and positing that the models crossed some kind of threshold of coherence around December 2025:
And then he followed it up by defining the term agentic engineering as this method of programming via LLM agents that is increasingly becoming the default workflow for professionals, to differentiate it from vibe coding, a term he coined a year ago:
Engineers who have been leveraging these models to code are able to do more and more and more, with multiple agents running 24/7 to build features and products — and hence why the feature backlog on the product roadmap may no longer be a backlog for a set of companies. At Footwork we’re floored on a weekly basis by the amount of product the tiny teams (1-5 people) we’re meeting with are able to build in a short period of time, and anecdotally it feels like there are more and more companies that fit this bill. SemiAnalysis published this week in Claude Code is the Inflection Point that 4% of GitHub public commits are being authored by Claude Code right now, and that will be 20% by year-end. In case you’re wondering what agentic engineering looks like IRL, OpenClaw founder Peter Steinberger posted a photo of his setup which features roughly 15 concurrent Codex sessions:
What does this mean?
There are four types of companies:
Those already living in the reality of seeing immense productivity gains from agentic engineering. The companies building the coding models themselves such as Anthropic, Cursor, and OpenAI are almost certainly the furthest along here.
Those who are learning what’s possible by already generating a lot of code. Many of them think they are agentic engineering, but they’re not anywhere close to the level of the above group of companies at the bleeding edge.
Those who are only just waking up to the coding models and their possibilities, by experimenting with them.
Those who have no idea about any of this.
Every company should be in the 2nd bucket after reading this post. The models are already showing productivity gains, and are only going to get better. Having your engineering team be as fluent as possible in these tools is table stakes. If you’re not leveraging the power of agents overnight, someone else competing with you is, and your product will be left behind. If you are in bucket 3, a post I found helpful from this week is by Mitchell Hashimoto, founder of HashiCorp: My AI Adoption Journey.
The good news for you as a startup is that many larger enterprises are in buckets 3 and 4. I’ve always believed the greatest advantages of a startup are speed and focus. AI is providing us with new capabilities on a daily basis to make speed an even greater superpower for those who are able to wield AI. Where to focus is still all about human judgment in most markets, though this is likely to become more agentic over time too.
Some have said that the broad selloff in public software companies in the past week - another “SaaSacre” - is effectively the public markets waking up to the power of Claude Code (as a proxy for agentic engineering); that is, how much software revenue may be at risk because software is becoming so easily developable, and thus commodified.
I think one of the keys for every public SaaS company this year is to show the market how capable they are of agentic engineering. Case studies of how quickly they are shipping new features and products leveraging code generation. Those new features and products driving revenue acceleration. More profitability by doing more with less. Enterprise AI ROI.
And for every company hoping to be public this year, wow does this feel important. Claude Code may be killing the feature backlog. Let’s hope it doesn’t kill the IPO backlog too.
I started next big thing to share unfiltered thoughts. I’d love your feedback, questions, and comments!
👇🏽 please hit the ♥️ button below if you enjoyed this post.