minware

2024 was the year of reckoning for AI. My most-read posts by far (50k and 20k impressions for #1/#2 vs. 5k for #3) were about how AI is only as good as a very junior engineer, and how we’d be better off automating easier tasks (like labeling Jira tickets) than hard ones (like building software). My friends and family range from total disdain (“I hate the tab suggestions in gmail — I know what I want to write!”) to extreme optimism (“AI can’t replace software engineers today, but just wait until next year!”) LLMs are unquestionably the most powerful advancement in computer science in the past decade, but are clouded by hype. People don’t understand how to use them yet, and many billions of venture capital are flushing down the drain in a wave of FOMO and euphoria. In 2025, winners will emerge who use this new tool well, and those running on hype capital with vague promises of replacing engineers will be seen for what they are. Which side are you on?

People talk about replacing engineers with AI — what about engineering managers? Managers identify anti-patterns (e.g., not reaching out for help soon enough), provide coaching, and track progress. This seems easier to do with AI than building software itself. Of course, AI can’t deal with politics or human problems like lack of motivation. But, it could do a lot of the job, especially for junior-to-mid-level engineers who face similar issues.

Tell ChatGPT to write a performance review based on what you know about your own work this year. If it’s better than your actual performance review from your manager, then it’s time to look for a new job.

The difference between a good and a great engineer is completing the same project with half the features and 90% of the value. Boards and CEOs who drive their teams to increase velocity miss this point: what you do is more important than how quickly you do it. (And if you don’t give your engineers a say in what you do, then you are beyond help…)

If you're worried about people "cheating" with LLMs during your interview process (beyond the initial screening), then your process is horribly broken to begin with.

Running a startup requires juggling an inhuman number of responsibilities. The best productivity hack I’ve found is *extreme single-tasking*. Scheduling focus time and going on do not disturb is great, but it’s not enough to avoid the eventual creep of interruptions and delays as an organization grows. Extreme single-tasking takes things to another level, structuring the entire organization so that everyone can complete their work from start to finish in one sitting. Achieving this level of productivity requires a deep commitment to single-tasking. This article talks about how to build an extreme single-tasking culture and offers advice about how to eliminate the most common causes of context switching. If you ever wanted to know how it’s possible for a small team to get more done than competitors several times their size, this is the secret.

The #1 benefit of AI copilots is that they force you to write better code (comments, variable names, etc.), otherwise they won’t work. Rewarding developers for writing down how their code should function by letting them type less is how we should be thinking about AI’s productivity impact. In a way, making the least experienced developers (and students) use it more may help them get better faster than directly writing code.

Researchers discover 9.5% of developers have no commits. Customers who’ve used a particular competitor tell us commits disappear due to squashes, rebases, and force pushes. Coincidence? 🤔 #dataquality

I hate having to talk to salespeople or submit my email so they’ll spam me. It’s like being forced into a car dealership before Tesla. Soon we will look back on SaaS companies who gate stuff behind email forms or don’t have full self-service product demos like e-commerce sites that used to make you create an account before checking out. Remember that?

VCs are pouring billions into AI writing code, while engineers spend hours filling out Jira tickets, tracking time, and writing status reports. Instead of replacing engineers, maybe first we should automate tedious non-engineering work that wastes their time.