Impact of AI on Human Programmers

A new study from METR shows that AI-assisted coding actually slowed down experienced developers by 19%, despite those same developers thinking they’d be 20–24% faster. This research actually reflects what I’ve seen firsthand with AI teams. Three takeaways stood out: 1️⃣ AI shines most with junior devs or new code, not complex legacy work. Use it where it adds real value, like onboarding, prototyping, boilerplate. 2️⃣ Perception ≠ reality. Devs believed they were faster even when they weren’t. So, always measure with hard KPIs, not just gut feeling. 3️⃣ Experience with the tool matters. Gains came only after 50+ hours with Cursor. You need training, domain context, and time to see the upside. Bottom line: AI is powerful if you use it in the right place, with the right team, and with process discipline. For code-heavy refactors, we’ll stay human-led… for now. Still, I’m bullish: with the right setup, senior teams will get faster. Curious how others are measuring actual impact from AI with mature teams. 👉 Full study here: https://lnkd.in/eWMNns2T

Everybody is talking about how AI will threaten developer jobs. But here's the thing: there are lots of people with a vested interest in having you believe that. 🤨 My take: all engineers will become AI engineers and there will be more software developers jobs than ever before. 🚀 Lots of people are motivated to have you believe that AI is magical and totally different from anything that existed before. Marketing teams who want a second chance for their product; CEOs who want regulators to slow down their competitors. At the end of the day, AI is just the newest platform. Like any new platform, AI supports powerful capabilities that did not exist before. And like all platforms before it, AI is going to empower a whole lot more developers. Whenever a transformational platform comes out, it enables people to build stuff they had no business building before. I grew up during the web revolution: HTML came out when I started kindergarten. Before I learned pre-algebra, I was able to build websites public to the entire world. Later in the 90s, people realizing they could build businesses on top of websites led to the dot-com boom. This led to a lot more demand for software developers than ever before. With this AI stuff, there’s the fear that making it easier to build software will reduce the need for software developers. Historically, this just hasn’t been true: platform shifts have led to new demand for new kinds of software, rather than fewer people building the same kinds of software. With the rise of the cloud, for instance, it no longer became necessary for companies to spend years building their own computing infrastructure. Because software shops no longer bottlenecked on the ability to hire compute and scaling experts, more businesses could become software businesses and demand for app developers increased more than ever. As the software industry shifts to AI-first, app engineers will need to become AI engineers. Organizations that previously bottlenecked on ability to create UIs, or to do CRUD programming, will now be able leverage AI. The bad news: a whole lot of today’s developers work on creating UIs or doing CRUD programming. The good news: it’s not hard for today’s skilled developers to become AI engineers. For many software developers, it’s true that your existing skills no longer give you job security. But a clear-thinking software developer who was building great products on mobile and cloud before can leverage their domain knowledge and software intuitions to build more impactful software even faster using AI. So far, we’re seeing AI empower more people to build more new kinds of software, instead of reducing the total number of people building the same software as before. This is great news for existing software developers. You already have experience with software and now there’s more of it to build! But buckle up; we’re in the middle of a whole lot of change. And after AI, we don’t even know yet what the next shift will be.

AI is taking all our jobs. Here's how I'm updating my resume. I'm only partly joking. I'm on the record: if you're in tech, AI isn't replacing you. From firsthand experience using AI, it's been a massive productivity boost. I wish I could clone myself to use even more AI. Nothing drove this home quite like teaching my kids to use AI. "Wow dad, you used to write all this code by hand? I just built a multiplayer game in 30 minutes and now my friends and I are playing it. Your job must have sucked." Ouch. But that's exactly what brings joy to software development. Sure, there are parts of coding that are a complete bore, but the art of creating software has always been awesome. It's never been more awesome than today, thanks to AI. I recently had the privilege of discussing AI's impact on tech with Scott Hanselman (check out his recent TEDxPortland), and it got me reflecting on what's actually changing. More AI: AI will get faster, cheaper, and gobble up more capabilities. As it becomes more efficient and accessible, we’ll apply it to more tasks, driving even greater demand for AI. It’s a modern echo of Jevons’ Paradox. More jobs (but different): Some roles will disappear, just as toll booth operators gave way to automation. But we’re not headed toward fewer opportunities, just different ones. As AI reshapes workflows, it shifts what needs doing and who does it. AI is fast, but also frustratingly slow: That first "this is amazing" moment quickly becomes "I could've just done this by now." Working with AI can feel like trying to get a child to follow instructions. Your clarity and context matter, and you’ll pay for being lazy with your prompts. AI won’t teach you what you don't know you don't know: Software was already a liability. Software built with AI? An even bigger liability. If you don’t understand the system, AI won’t save you, but it might help you ship broken things faster. The visual advantage is real: Using AI to generate applications you can see and interact with is a game-changer for builders. The visual feedback lets you verify correctness quickly. But backend code? That’s a different beast; harder to inspect, harder to trust. This is why we’re building Postman Flows: a visual, low-code, AI-native tool that lets builders see and verify how their applications work on the backend. We’ve used Flows internally to deploy dozens of AI-powered applications (agents), from turning Slack threads into Jira action plans, to handling product feedback. More and more, applications are just AI orchestrating APIs. Lots of APIs. Flows makes that explicit and buildable. Using AI-ready APIs (which Postman helps you define, test, and structure) is key to making this work reliably. These agents have already saved us hundreds of hours. We’re sharing the agents soon, so you can use them too. So here's how I'm updating my resume: spending all my time learning and building new things with AI. The future belongs to those who learn to work with AI.

AI was supposed to make us 24% faster. It made top developers 19% slower. That’s not a typo. That’s from a real world study. So what actually slows people down? 1. Tool overload. Developers had to manage too many AI outputs. AI tools struggled with code context. Humans still fixed minor issues AI couldn’t catch. 2. Mismatched expectations. Developers expected 24% speed-up. Reported feeling faster even when slower. Believed AI helped after the fact. There’s a 43% gap between perception and reality. 3. Unrealistic testing environments. Real work is messy, full of edge cases. AI still struggles with multi-step real-world issues. This study used live coding tests on real open source codebases. 4. Short-term adoption, long-term learning curve. Developers only used AI tools for a few dozen hours. Most weren’t optimized power users. Higher returns may come after 100+ hours. Speed-up might emerge with sustained usage and better prompting. This doesn’t mean AI is useless. But it proves that evaluating AI’s real impact needs more than benchmark bragging rights. Want to know what makes AI truly work inside a team? What do you think drives real AI productivity gains?

AI impact to software engineering jobs Anecdotes are great but what does the data say? Analyze 20M job postings over 16 month period and some interesting insights come out (https://lnkd.in/eka32Kma) 1. AI Scientists and ML engineer demand is growing. Biggest losers are front end engineers and data engineers!. Data Scientists have been fairly resilient 2. Salary ranges however, are flat adjusting for inflation; and given the supply - likely that the software engineering will not see massive jumps in near term 3. NLP is the most desired skillset with anything LLM related going through the roof- chatbot anyone! 4. Rust is gaining ground, React is taking share from Angular for front end. Python remains resilient as the de facto ML programming language 5. And while, large tech companies which did massive laying offs of software engineers are hiring back, there is no evidence that they are hiring more AI engineers than others. They are just hiring across the board. Maybe they are upgrading talent and cleaning up from the bad hiring practices of 2021 The headline is however clear: If you are graduating to become a software engineer or an existing software engineer; you will do well to add more AI skills. At least, that is what the data says. 

The conversation around AI and jobs often feels hypothetical—but for programmers, it’s already reality. In my latest article, “GenAI Is Reducing Demand for Programmers. It Won’t Stop There,” I dive into the numbers: Programmer roles declined 26% in just two years—a trend that accelerated with the rise of GenAI. Related roles like web development, support, and system administrator are showing early signs of similar pressure. On the other hand, some roles have seen strong growth: InfoSec Analyst jobs grew by 36% in this same period. This isn’t just a tech story—it’s a talent strategy challenge. At Kelly SETT, we’re helping organizations and professionals prepare for a future where automation reshapes roles faster than ever before. Read the full analysis below. What do you think—are companies ready for this shift? #FutureOfWork #AI #KellySETT #TechTalent #Leadership