I believed I’d provide a couple of takeaways and reflections based mostly on final week’s first AI Codecon digital convention, Coding with AI: The Finish of Software program Growth as We Know It. I’m additionally going to incorporate a couple of brief video excerpts from the occasion. When you registered for Coding with AI or for those who’re an current O’Reilly subscriber, you’ll be able to watch or rewatch the entire thing on the O’Reilly studying platform. When you aren’t a subscriber but, it’s straightforward to begin a free trial. We’ll even be posting extra excerpts on the O’Reilly YouTube channel within the subsequent few weeks.
However on to the promised takeaways.
First off, Harper Reed is a mad genius who made everybody’s head explode. (Camille Fournier apparently has joked that Harper has rotted his mind with AI, and Harper truly agreed.) Harper mentioned his design course of in a chat that you simply may wish to run at half velocity. His greenfield workflow is to start out with an thought. Give your thought to a chat mannequin and have it ask you questions with sure/no solutions. Have it extract all of the concepts. That turns into your spec or PRD. Use the spec as enter to a reasoning mannequin and have it generate a plan; then feed that plan into a special reasoning mannequin and have it generate prompts for code technology for each the appliance and checks. He’s having a wild time.
Agile Manifesto coauthor Kent Beck was additionally on Workforce Enthusiasm. He advised us that augmented coding with AI was “essentially the most enjoyable I’ve ever had,” and mentioned that it “reawakened the enjoyment of programming.” Nikola Balic agreed: “As Kent mentioned, it simply introduced the enjoyment of writing code, the enjoyment of programming, it introduced it again. So I’m now producing extra code than ever. I’ve, like, one million traces of code within the final month. I’m enjoying with stuff that I by no means performed with earlier than. And I’m simply spending an obscene quantity of tokens.” However sooner or later, “I believe that we gained’t write code anymore. We’ll nurture it. This can be a imaginative and prescient. I’m certain that lots of you’ll disagree however let’s look years sooner or later and the way all the pieces will change. I believe that we’re extra going towards intention-driven programming.”
Others, like Chelsea Troy, Chip Huyen, swyx, Birgitta Böckeler, and Gergely Orosz weren’t so certain. Don’t get me incorrect. They assume that there’s a ton of wonderful stuff to do and study. However there’s additionally lots of hype and unfastened considering. And whereas there will probably be lots of change, lots of current expertise will stay essential.
Right here’s Chelsea’s critique of the current paper that claimed a 26% productiveness improve for builders utilizing generative AI.
If Chelsea will do a sermon each week within the Church of Don’t Consider The whole lot You Learn that consists of her displaying off numerous papers and giving her dry and insightful perspective on how to consider them extra clearly, I’m so there.
I used to be a bit stunned by how skeptical Chip Huyen and swyx have been about A2A. They actually schooled me on the notion that the way forward for brokers is in direct AI-to-AI interactions. I’ve been of the opinion that having an AI agent work the user-facing interface of a distant web site is a throwback to display scraping—absolutely a transitional stage—and whereas calling an API will probably be one of the simplest ways to deal with a deterministic course of like cost, there will probably be an entire lot of different actions, like style matching, that are perfect for LLM to LLM. Once I take into consideration AI purchasing for instance, I think about an agent that has realized and remembered my tastes and preferences and particular targets speaking with an agent that is aware of and understands the stock of a service provider. However swyx and Chip weren’t shopping for it, no less than not now. They assume that’s a good distance off, given the present state of AI engineering. I used to be glad to have them deliver me again to earth.
(For what it’s price, Gabriela de Queiroz, director of AI at Microsoft, agrees. On her episode of O’Reilly’s Generative AI within the Actual World podcast, she mentioned, “When you assume we’re near AGI, attempt constructing an agent, and also you’ll see how far we’re from AGI.”)
Angie Jones, alternatively, was fairly enthusiastic about brokers in her lightning discuss about how MCP is bringing the “mashup” period again to life. I used to be struck particularly by Angie’s feedback about MCP as a type of common adapter, which abstracts away the underlying particulars of APIs, instruments, and knowledge sources. That was a strong echo of Microsoft’s platform dominance within the Home windows period, which in some ways started with the Win32 API, which abstracted away all of the underlying {hardware} such that utility writers not needed to write drivers for disk drives, printers, screens, or communications ports. I’d name {that a} energy transfer by Anthropic, apart from the blessing that they launched MCP as an open commonplace. Good for them!
Birgitta Böckeler talked frankly about how LLMs helped cut back cognitive load and helped assume by way of a design. However a lot of our every day work is a poor match for AI: massive legacy codebases the place we alter extra code than we create, antiquated know-how stacks, poor suggestions loops. We nonetheless want code that’s easy and modular—that’s simpler for LLMs to grasp, in addition to people. We nonetheless want good suggestions loops that present us whether or not code is working (echoing Harper). We nonetheless want logical, analytical, essential desirous about drawback fixing. On the finish, she summarized each poles of the convention, saying we’d like cultures that reward each experimentation and skepticism.
Gergely Orosz weighed in on the continued significance of software program engineering. He talked briefly about books he was studying, beginning with Chip Huyen’s AI Engineering, however maybe the extra essential level got here a bit later: He held up a number of software program engineering classics, together with The Legendary Man-Month and Code Full. These books are many years outdated, Gergely famous, however even with 50 years of device improvement, the issues they describe are nonetheless with us. AI isn’t prone to change that.
On this regard, I used to be struck by Camille Fournier’s assertion that managers like to see their senior builders utilizing AI instruments, as a result of they’ve the talents and judgment to get essentially the most out of it, however usually wish to take it away from junior builders who can use it too uncritically. Addy Osmani expressed the priority that fundamental expertise (“muscle reminiscence”) would degrade, each for junior and senior software program builders. (Juniors could by no means develop these expertise within the first place.) Addy’s remark was echoed by many others. No matter the way forward for computing holds, we nonetheless must know analyze an issue, how to consider knowledge and knowledge buildings, design, and debug.
In that very same dialogue, Maxi Ferreira and Avi Flombaum introduced up the critique that LLMs will have a tendency to decide on the commonest languages and frameworks when making an attempt to unravel an issue, even when there are higher instruments out there. This can be a variation of the remark that LLMs by default have a tendency to supply a consensus resolution. However the dialogue highlighted for me that this represents a threat to talent acquisition and studying of up-and-coming builders too. It additionally made me marvel about the way forward for programming languages. Why develop new languages if there’s by no means going to be sufficient coaching knowledge for LLMs to make use of them?
Virtually all the audio system talked in regards to the significance of up-front design when programming with AI. Harper Reed mentioned that this appears like a return to waterfall, besides that the cycle is so quick. Clay Shirky as soon as noticed that waterfall improvement “quantities to a pledge by all events to not study something whereas doing the precise work,” and that failure to study whereas doing has hampered numerous initiatives. But when AI codegen is waterfall with a quick studying cycle, that’s a really totally different mannequin. So this is a crucial thread to tug on.
Lili Jiang’s closing emphasis that evals are rather more advanced with LLMs actually resonated for me, and was in line with lots of the audio system’ takes about how a lot additional we’ve to go. Lili in contrast an information science challenge she had accomplished at Quora, the place they began with a fastidiously curated dataset (which made eval comparatively straightforward), with making an attempt to take care of self-driving algorithms at Waymo, the place you don’t begin out with “floor reality” and the precise reply is very context dependent. She requested, “How do you consider an LLM given such a excessive diploma of freedom when it comes to its output?” and identified that the code to do evals correctly may be as massive or bigger than the code used to form the precise performance.
This completely suits with my sense of why anybody imagining a programmer-free future is out of contact. AI makes some issues that was onerous trivially straightforward and a few issues that was straightforward a lot, a lot more durable. Even for those who had an LLM as decide doing the evals, there’s an terrible lot to be found out.
I wish to end with Kent Beck’s considerate perspective on how totally different mindsets are wanted at totally different levels within the evolution of a brand new market.
Lastly, an enormous THANK YOU to everybody who gave their time to be a part of our first AI Codecon occasion. Addy Osmani, you have been the proper cohost. You’re educated, a terrific interviewer, charming, and lots of enjoyable to work with. Gergely Orosz, Kent Beck, Camille Fournier, Avi Flombaum, Maxi Ferreira, Harper Reed, Jay Parikh, Birgitta Böckeler, Angie Jones, Craig McLuckie, Patty O’Callaghan, Chip Huyen, swyx Wang, Andrew Stellman, Iyanuoluwa Ajao, Nikola Balic, Brett Smith, Chelsea Troy, Lili Jiang—you all rocked. Thanks a lot for sharing your experience. Melissa Duffield, Julie Baron, Lisa LaRew, Keith Thompson, Yasmina Greco, Derek Hakim, Sasha Divitkina, and everybody else at O’Reilly who helped deliver AI Codecon to life, thanks for all of the work you place in to make the occasion successful. And because of the virtually 9,000 attendees who gave your time, your consideration, and your provocative questions within the chat.
Subscribe to our YouTube channel to observe highlights from the occasion or develop into an O’Reilly member to observe your complete convention earlier than the following one September 9. We’d love to listen to what landed for you—tell us within the feedback.