The automated setup improved machine learning inference speeds from an initial baseline of 6 tokens per second to 213 tokens per second. A cloud agent operating inside Conductor leveraged Claude and OpenAI models, including Astra and Opus, to complete the optimization. The developer exercised minimal oversight, providing occasional machine learning suggestions while lacking a background in Verilog and holding little recent C++ experience. Over hundreds of experiments, the project burned through around $10,000 in tokens while generating millions of lines of code.
Key points
Automated coding agents can independently drive significant hardware-level optimizations,…: Automated coding agents can independently drive significant hardware-level optimizations, boosting model processing speeds from 6 up to 213 tokens per second.
Complex hardware programming can be achieved by developers lacking specific domain expert…: Complex hardware programming can be achieved by developers lacking specific domain expertise in specialized languages like Verilog.
Strategic guidance was minimal, consisting largely of high-level machine learning advice…: Strategic guidance was minimal, consisting largely of high-level machine learning advice and one summarized prompt from DeepGrove's founders at the 10 tokens per second mark.
Sources checked by our analysis pipeline: 1
One of the more entertaining things about programming recently is that coding agents let me work in entirely new domains by giving me access to new programming languages and paradigms. Lately, that has meant hacking on hardware, now that writing low-level software is much easier. So, in this blog post, I’m going to try to recount the experience of vibe coding an FPGA, going through a series of “oh shit” moments several times, and giving myself a new appreciation for what’s possible if you have a large token budget (thanks YC) and interesting problems to point the models at.
I overheard some founders recently talk about how FPGAs work. I’ll admit I was relatively naive about them. I’d never really experimented with them in my startup or in college, so I knew little about them before. The basic idea I was familiar with: you can program a circuit into a reprogrammable chip. So rather than having a general purpose computing architecture (like a CPU) you can have a reprogrammable one, i.e. the configuration loaded onto the chip determines how its…
Unconstrained automated coding assistants frequently make flawed architectural choices that severely degrade software performance while blindly fulfilling functional requirements. To prevent bloated systems, engineers are now forced to shift away from hands-on programming to establish strict boundaries and review underlying data structures. This sudden evolution in responsibilities creates friction as technical teams must actively manage tool outputs rather than writing software.
Evaluation nonprofit METR has revised its widely cited tracking metric, revealing that top artificial intelligence models struggle significantly with open-ended tasks requiring subjective human discernment. While systems continue to rapidly improve in easily checkable fields like mathematics and coding, current reinforcement learning methods fail to instill broader judgment. This persistent shortcoming challenges assumptions about how quickly automated tools can independently take over advanced research and develo…
According to reporting from the Wall Street Journal, an archived news aggregation page documented a public challenge regarding illicit technology transfers. The matter centers on assertions from congressional intelligence leadership that critical artificial intelligence hardware warnings were disregarded. The available archival record provides no surrounding context or responses from the parties involved.