Sonja Petrović Lundberg

was at OpenAI hq yesterday and got to talk with their head of product about the new agentkit 🤯 a lot of people think this kills zapier (i even thought so at first). it doesn’t. agentkit is for devs, not no-coders. it’s about putting ai inside products, not automating workflows. this makes it easier for teams to build agents that live inside their own apps, not connect tools together. so yeah, this isn’t about replacing zapier. it’s about replacing ui. #ai #openai #chatgpt #agents #startups #developers

124

8 kommentarer

I caught up with a friend who works at a mid-size Swedish tech company. Over the last 4 months, their shipping velocity has almost doubled – not because they hired more engineers, adopted some new agile framework, or worked late nights. It came down to a single change in how they build products: they started using Lovable to prototype features instead of writing traditional spec docs. Before Lovable, it usually went like this: PMs drafted long PRDs, trying to anticipate every detail. Multiple stakeholders reviewed these documents, leaving comments and raising concerns. The document grew with each iteration. Alignment meetings were frequent but often resulting in ambiguity. Engineers often began implementation while details were still debated. Inevitably, confusion emerged about trade-offs, timelines got pushed, and features shipped incomplete or scaled back. Now, PMs build interactive prototypes directly in Lovable. These aren’t wireframes or rough mockups – they’re fully clickable, end-to-end experiences that feel like the real product. Engineers don’t have to guess what the flow should be. Designers don’t have to explain interactions. Everyone sees the same thing, from day one. The end result is fewer meetings, fewer misunderstandings, fewer rewrites. What used to take weeks of coordination now happens in a single day. This is what has provided the most value for enterprises using Lovable so far. Over time, the increase of clarity and velocity saves the companies millions of $ in wasted effort.

2 480

175 kommentarer

Lots of people keep asking me, “How do I build a moat in today's AI-everywhere world?” My simple answer: speed of execution. OpenAI took Codex from idea to public product in seven weeks, eight engineers, zero quarterly‑planning detours. When Replit's AI agent accidentally nuked a customer’s production database, the team pulled an all‑nighter, restored the data, and shipped new guardrails, all inside 24 hours. On the funding side, Dharmesh Shah jumped on a Sunday Zoom with Lovable’s founder Anton Osika and wired the check first thing Monday. Even the giants get it: Google and Meta snap up top talent in 48‑72 hours, whatever it takes. For every success story, there’s another team killing an idea faster than slower teams can finish their pitch decks. That’s product velocity, deal‑making velocity, hiring velocity, compounding while others complain that “AI is too crowded.” At Haptik, we had a rule: "inaction is worse than failure". In 2025, inaction isn’t just worse, it’s lethal. If your AI idea needs 4-5 months of “alignment etc.” before users touch it, someone faster has already eaten your lunch. So let’s build. Drop the overthinking, quit waiting for approval, and ship something today. Then learn, iterate, and ship again.

103

3 kommentarer

50% of new code at Robinhood is AI-Generated But do you actually see that level of visibility in your own organization? Half of Robinhood’s new code is already written by AI. (Business Insider) That’s not a futuristic projection, it’s happening right now in production environments. The reality is that most engineering leaders don’t know how much of their codebase is AI-generated, which teams rely on it the most, or whether it’s improving delivery speed without adding hidden debt. They’re left with assumptions instead of evidence. This blind spot matters. Without clarity, budgets are allocated to tools that may never get adopted, while critical questions remain unanswered: Is AI shortening cycle times? Is it raising bug density? Is the ROI visible to finance, not just engineering? GenAI is reshaping engineering output at scale. At Milestone we provide measurable insights into AI-generated code, adoption across teams, and the real ROI of tools like Copilot. So the next time someone asks how much of your code is written by AI, you’ll have more than a guess.

251

3 kommentarer

It’s clear that almost everything about software development has fundamentally changed for good with AI agents. Hard to imagine another area of knowledge work that has had its underlying process get transformed so substantially in such a short period of time. If you went back even 2 years and told the average developer what software engineering would look like right now, it would not be comprehensible. We’ve already seen 3-4 major step functions changing workflows just in that time period alone. The ultimate outcome is clearly that software engineering will become heavily dependent on how good you are at prompting, how good you are at reviewing code, and the interplay between the two. Once the workflow becomes about kicking off agents to do work in the background, your success is dependent on the work that gets kicked off and how you review it. Importantly, both require a deep understanding of how to code, systems, and what you want to build; the greater the underlying skill of the engineer, the greater the leverage with AI agents. And while we’re seeing this workflow get defined first in software, a similar pattern will eventually emerge across all areas of knowledge work. We’re only in the earliest stages of seeing what this will look like. Crazy times.

219

43 kommentarer

Today, I’m excited to bring GeoLift by Recast to the whole world. After months of positive feedback from clients and partners, we’re opening up Recast’s incrementality testing platform as a standalone product. And we’re giving it away, completely free, for 6 months. Why? Our core goal at Recast has always been to make good causal inference methods available to more marketers. GeoLift by Recast does this with a simple, yet powerful platform for designing and analyzing geographic-based lift tests. Here are some of the things you can do with it: ✓ Design clear, rigorous incrementality tests in <15 minutes ✓ Measure the incremental impact of online and offline media channels ✓ Run unlimited tests – no PhD’s or complex setups required GeoLift by Recast meets the same standards of statistical rigor you’ll find with any Recast product and you can read more about the methodology in our extensive documentation. To get started, follow the link in the comments to create an account and start designing your first incrementality test!

273

34 kommentarer

Interesting visual of a head-of-line blocking behavior in the Moderne CLI before and after a fix. Each block here represents a different repository with an #OpenRewrite recipe running on it in a large-scale multi-repository run. Knut Wannheden found the bug here yesterday and fixed. Concurrency is hard and the visuals here were produced by Claude Code looking at Moderne trace data we co-designed with our friends at American Express and introduced in 3.45.0. Next, we're exploring different prioritization methods where we're trying to balance between two apparently competing objectives: 1. Runs that should be fast are prioritized early to show incremental results as quickly as possible. After all, if a recipe result is incorrect on the first few repositories, developers like to cancel them quickly and iterate further on the recipe. 2. Runs on repositories that are going to take the longest shouldn't be prioritized last or we have the long tail effect on a run that you can observe even after the fix to the head-of-line blocking bug. We're thinking that Highest Response Ratio Next (HRRN) is a good balance between the two with a simple formula shown below. Over the years we've iterated on a "weight" measure for repositories based roughly on the density of information (including type attribution) of a repository present in the LST. Regardless of which method we wind up choosing, we'll be using this weight measure as one that run time is roughly proportional to. I know it may be a little outside of your typical problem space Geoffrey De Smet, but do you have any suggestions? I imagine our problem area is quite a bit simpler than yours typically. 😉

46

5 kommentarer

Today, we're excited to drop the first episode of our new podcast - 𝘏𝘰𝘸 𝘞𝘦 𝘉𝘶𝘪𝘭𝘵 𝘐𝘵. When engineering and design work together, truly magical things happen. At Clueso, that collaboration is at the heart of everything we do, because we’re literally building a design tool to help people create magical designs of their own. In this series, we’re pulling back the curtain on how we build at Clueso (YC W23). We'll uncover the experiments we run, the late-night breakthroughs, and the countless little details that turn our ideas into the products we build. Our first episode dives into how we launched expressive voiceovers in Clueso, powered by the ElevenLabs v3 API. Within hours of ElevenLabs announcing the update, our customers began asking when it would be available in Clueso. By the next day, it was live in production, complete with a full launch video. This is the story of how we made that happen. Watch the full episode on YouTube here: https://lnkd.in/gq5Jzs98 Had a great time chatting with Prajwal Prakash and Ajinkya Gajbhiye.

102

By Q2 2025, 33% of organizations had already deployed AI agents into production, up from 11% just months earlier. If those agents are running live systems without proper guardrails, every misconfiguration becomes a business risk. The problem is already visible. In July 2025, 13% of organizations reported breaches of AI models or applications, and 97% admitted they lacked proper AI access controls. This is a production incident caused by missing discipline. At the same time, 98% of enterprises plan to expand their use of AI agents, while 96% of them acknowledge these agents are becoming a major security threat. Leaders know the exposure is growing, but adoption races ahead anyway. The fix isn’t complex: ֻ• give every agent a unique identity, lock permissions to scope, and enforce immutable logs. • Validate every action against the agent’s role before execution. Without this, your agents aren’t accelerating work, they’re multiplying your blast radius. The real question is whether you can map which agents have access to what today and whether you could untangle that picture during an incident.

387

6 kommentarer