I thought I would present some fast food and repercussions based on the first virtual conference of AI Codecon last week, Costume with artificial intelligence: the end of software development as we know. I will also include some short video excerpts from the event. If you register for coding with artificial intelligence or if you are currently subscribed to O’Railly, you can watch or re -watch everything on the O’Reillly learning platform. If you are not subscribed yet, it is easy to do so Start a free experience. We will also post additional excerpts on Orili YouTube channel In the next few weeks.
But to the promised fast food.
First, Harper Red is a crazy genius that made everyone explode. (It seems that Cameel Furnier has joined that Harper has crushed his brain with artificial intelligence, Harper actually agreed.) The course of his work in Greenfield is the start of an idea. Give your idea to the chat model and ask you about questions with yes/no answers. Make him extract all ideas. This becomes specifications or PRD. Use specifications as inputs of the thinking model and make them create a plan; Then feed this plan in a different thinking model and make it create claims to generate software instructions for both the application and tests. It has wild time.
https://www.youtube.com/watch?
A slim statement The co -author Kent Beck was also in the team’s enthusiasm. He told us that a strengthened coding with artificial intelligence was “more enjoyable at all”, and said he “re -woke up the joy of programming.” Nicolas Balik agreed to: “As Kent said, I brought the joy of the writing code, the joy of programming, I brought it back. So I am now creating a symbol more than ever. Like, like, a million lines of the code last month. But in the future, “I think we will not write a symbol anymore. We will nurture it. This is a vision. I am sure that many of you will differ, but let’s look at years in the future and how everything will change. I think we are heading towards the intentional programming.”
Others, such as Chelsea Troy, Shayb Hoin, Swix, Bergita Boukiller, and Jerevian Oras are not sure. Do not understand me wrong. They think there are a lot of amazing things to do and learn. But there is also a lot of noise and loose thinking. Although there will be a lot of change, many current skills will remain important.
Here Chelsea’s criticism of The modern paper that has claimed 26 % production increased For developers using obstetric artificial intelligence.
https://www.youtube.com/watch?
If Chelsea will make a sermon every week in an incredible church, everything you read consists of displaying different papers and giving them a dry and insight into how to think more clearly, then I am.
I was a little surprised how CHIP HUYEN and SWYX skeptical about A2A. He truly studied me in the idea that the future of agents in the direct organization of Amnesty International. I have seen that the presence of Amnesty International’s agent is working, the interface facing a user for the distant website is a bouncing to scan-scanning-transition, and while calling an application programming interface will be the best way to deal with an inevitable process such as payment, there will be a large group of other activities, such as matching taste, which is ideal for LLM to LLM. When I think about shopping artificial intelligence, for example, I imagine an agent learning, remembering my taste, preferences, and specific goals in communicating with an agent who knows and understands the merchant inventory. But Swyx and Chip did not buy it, at least not now. They think this is far from a state, given the current state of artificial intelligence engineering. I was happy to return them to the ground.
(What deserves, Gabriella de Kerez, the director of artificial intelligence at Microsoft. Episode From Aurelli Ai Toulidi in the real world She said, “If you think we are close to AGI, try to build an agent, and you will see to what extent we are from AGI.”
Angie Jones, on the other hand, was very excited about the agents Modernity On how MCP re -erad “Mashup” to life. In particular, I was shocked by Angie’s comments on MCP as a kind of comprehensive transformer, which raises the basic details of applications, tools and data sources. This was a strong resonance to dominate the Microsoft platform in the Windows era, which started in many ways with the WIN32 Application interface, which extracted all the basic devices that the application book is no longer to write drivers driving drivers, printers, screens or communications outlets. I would like to call this power on the anthropoor, with the exception of the pool that MCP presented as an open standard. Good for them!
Birgitta Böckeler spoke frankly about how LLMS helped reduce cognitive pregnancy and helped think through design. But a lot of our daily work is suitable for artificial intelligence: the rules of old old symbols where we change a more symbol than we create old technology chimneys and weak feedback rings. We still need a simple and standard symbol – this is easier for llms to understand, as well as humans. We still need good reactions that show us whether the code is working (Echoes Harper). We still need logical, analytical and critical thinking about solving problems. In the end, both poles were summarized from the conference, saying that we need cultures equivalent to both experimentation and suspicion.
Geerly Orosz weighs the importance of continuous software engineering. Shortly talk about the books he was reading, starting with Chip Huyen’s Artificial Intelligence EngineeringBut perhaps the most important point was just: he carried many software engineering classes, including The legendary man’s monthly and Full symbol. These books are decades old, I noticed Girley, but even with 50 years of developing tools, the problems they describe with us are still with us. Artificial intelligence is unlikely to change this.
In this regard, Camille Furnier surprised me that managers like to see the great developers who use artificial intelligence tools, because they have skills and judgment to achieve the maximum benefit from it, but they often want to take it from the novice developers who can use them unusual. Addy Osmani expressed anxiety that the basic skills (“muscle memory”) will turn, whether for novice or senior software developers. (Young children may never develop these skills in the first place.) Many others have repeated ADDY. Regardless of the future of computing, we still need to know how to analyze the problem, how to think about data structures, data structures, how to design, and how to correct.
In the same discussion, MAXI Ferreira and AVI FLOMBAUM categorization that LLMS will tend to choose the most common languages and frameworks when trying to solve a problem, even when there are better tools. This is a variation in the observation that LLMS is default to produce a consensus solution. But the debate highlighted for me that this is a danger to the acquisition of skills and the learning of new developers as well. It also made me wonder about the future of programming languages. Why do new languages evolve if there is not enough LLMS training data to use?
Almost all speakers talked about the importance of the design presented in programming with artificial intelligence. Harper Reed said this looks like a return to the waterfall, but the cycle is very fast. Clay Skiller Once observed The development of the waterfall “amounts to a pledge by all parties not to learn anything while doing actual work”, and this failure in learning while doing it has hindered countless projects. But if Codegen AI is the waterfall with a quick learning cycle, then this is a completely different model. So this is an important topic that is withdrawn.
Lily Jiang’s closing focus on Evals is really more complicated with LLMS’s echo for me, and he was consistent with many speakers receiving how much we should go to. Lily compared her data science project in Quora, where she started with a carefully coordinated data set (which made Eval relatively easy), with an attempt to deal with the self -driving algorithms in Waymo, as it does not start with the “terrestrial truth” and the correct answer depends on the very context. I asked, “How do you evaluate LLM that gave this high degree of freedom in terms of its production?” He pointed out that the code that must be done correctly can be large or larger than the code used to form the actual function.
This fits perfectly with my feeling of the reason that makes anyone imagine a program free of the programmer away from touch. Artificial intelligence makes some things that were very easy and some things that were easy and much difficult. Even if you have LLM as a judge who does Evals, there are many things to be discovered.
I want to finish the studied Kent Beck perspective on how to need a different mentality in different stages in the development of a new market.
https://www.youtube.com/watch?
Finally, thank you very much to everyone who gave their time to be part of the first Codecon Ai event. Addy Osmani, you are the perfect Cohost. You are familiar with, great interview, charming, and a lot of fun to work with it. Geergely 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, and Rewrewman, Brett Smith, Chelsea Troy, Lily Jiang – all of them were defeated. Thank you very much for sharing your experience. Melissa Dofeld, Julie Baron, Lisa Lario, Keith Tombson, Yasminina Greco, Derek Hakim, Sasha Devikina, and everyone else in Urieli who helped bring Codecon ai, thanks for all the work that I put in this event success. Thanks to approximately 9,000 of those present who gave your time, attention and provocative questions in the chat.
Subscribe to Our YouTube channel To see the prominent points of the event or Be a member of O’Railly To see the entire conference before the next one on September 9. We would like to hear what fell to you – let’s know in the comments.
