In creating the content material for our Could 8 digital convention Coding with AI: The Finish of Software program Growth as We Know It, we couldn’t assist however need to function Harper Reed, whose latest publish “My LLM Codegen Workflow ATM” so completely encapsulates the sort of experimentation that builders are going by as they arrive to grips with the transformation that AI is bringing to how they work, what they will accomplish, and which instruments they need to be adopting. Harper lays out his present workflows and instruments with detailed examples for each greenfield code and legacy code that make it straightforward for others to study from what he’s accomplished. It’s a terrific mannequin for the sort of info sharing we hope to engender with the upcoming occasion and others that may observe. For those who haven’t learn it but, go accomplish that now.
In an e-mail to me, my previous good friend Nat Torkington had this to say about Harper’s publish:
Be taught quicker. Dig deeper. See farther.
I really feel like there are ascending ranges of nerd on this:
– immediate hacks
– instruments to combine into your workflow
– context hacks (e.g., construct a necessities doc with the LLM, then get it to code to these necessities)
– use of particular fashions/options (e.g., reasoning vs. non-reasoning LLMs)
– {custom} workflows/instruments assembled from items, however which actually are custom-built to that particular person’s workflow
ALL of those are issues that we need to cowl in our upcoming occasion. So I’d like to make use of Harper’s piece as a immediate and context hack to all of you, to make clear what we’re searching for on the occasion. Coding with AI will function fireplace chats with of us like Jay Parikh, Addy Osmani and Gergely Orosz, Chip Huyen, and Shawn @swyx Wang, and talks by Harper, Simon Willison, Chelsea Troy, Steve Yegge, Andrew Stellman, and others. However we’re nonetheless searching for extra studies from the trenches, offered as five-minute lightning talks (a format Nat Torkington initially developed for our Perl Convention practically 30 years in the past, and that was a beloved function of all our conferences thereafter).
Classes we’re fascinated with embrace “my favourite AI software,” “my favourite AI immediate or context hack,” “my workflow,” “my venture that I might by no means have thought to attempt with out the benefit of experiment I get with AI,” and so forth. Along with the lightning talks at this occasion, we’re additionally planning a future recurring occasion that’s solely dedicated to a stay model of the sort of full show-and-tell that Harper did so properly.
So, go learn Harper, and present us what you’ve obtained! Ideally, you’ll not solely give us the total model but additionally distill specific, usable classes from it. Right here’s some further knowledge from Nat’s e-mail to get you began:
The extra the mechasuit is designed on your cortex, the much less transferable it’s to different individuals. The danger is you begin with “listed below are six instruments I take advantage of in my very own complicated workflow in a language you don’t recognise to resolve issues you don’t have” and lose individuals in a flood of random software names.
[We want to be] ready to have a look at totally different individuals’s “I code with AI!” workflows and break them down [as Harper does] so [our viewers] can go “properly, that’s cool however I’m not utilizing vi and command-line instruments don’t play properly with Visible Studio, so I’ll simply steal your context hacks for working with large code bases in Copilot.”
Addy Osmani added to our e-mail dialog:
Constructing on the “ascending ranges of nerd” framework (which I really like!), we may construction a number of the “how I work” content material to particularly tackle totally different developer expertise ranges. For instance:
- Junior devs: concentrate on constructing that important analysis mindset and understanding when/easy methods to successfully use AI instruments
- Mid-level: exploring workflow integration and context optimization
- Senior: deep dives into {custom} tooling and superior immediate engineering
Submit your proposals (together with, ideally, a publish modeled on Harper’s that we will publish right here on the O’Reilly Radar weblog) on the present name for displays hyperlink. We’ve up to date the submission deadline to March 12 and the occasion date from April 24 to Could 8 to offer you a bit extra time to do your reasoning after which reply to this revised immediate.
On Could 8, O’Reilly Media will probably be internet hosting Coding with AI: The Finish of Software program Growth as We Know It—a stay digital tech convention spotlighting how AI is already supercharging builders, boosting productiveness, and offering actual worth to their organizations. For those who’re within the trenches constructing tomorrow’s improvement practices at the moment and fascinated with talking on the occasion, we’d love to listen to from you by March 12. You’ll find extra info and our name for displays right here.