Luiz Fernando Scheidegger

We’re all familiar with first-mover advantage. But is AI quietly making second-mover advantage more powerful? Dalton Wais and I have been talking a lot about how our roles as engineers have shifted over the last few months. AI is pushing engineers further upstream, closer to product management. When you can build almost anything, what you choose to build matters more than ever. That was always true, but it feels amplified now. A lot of our time recently has gone into experimenting with how we define, analyze, and capture product requirements. Some of this is still very traditional. Talking to our ICP, listening carefully, synthesizing notes, analyzing transcripts. But more and more, we’re using AI to surface requirements at a depth and speed that just wasn’t possible before. For the payroll system I’m working on, that looks like a few things. Asking AI to find and analyze every payroll product demo video on YouTube, not just reading transcripts but inspecting the UI itself frame by frame, and producing a comparative matrix of features across vendors. Asking AI to scan blogs, Reddit, and forums for people complaining about payroll software, clustering those complaints, and turning them into candidate requirements. Asking AI to read thousands of reviews across G2 and similar sites and separate what customers say they want from what they say they hate. And then layering all of that together. This is what got me thinking about second-mover advantage. When you’re the first mover, very little of this data exists. The market is mostly theory and opinion. But once a category matures, it generates an enormous exhaust of demos, reviews, complaints, and expectations. AI can now consume that exhaust almost instantly. If you can take a well-understood set of requirements and turn it into working software quickly, the old moat of “we’ve been building this for years” starts to matter less. Time to market was a real advantage pre-AI. Now time to insight might be the bigger one. Of course, this just pushes pressure onto the other hard problems. Narrative. Distribution. Lock-in. First movers still have advantages there, if they execute well. But the quiet moat of time spent just building is eroding fast.

10

DeepSeek’s rise has been wild to watch. From hedge fund roots to topping app store charts, all while running models cheaper and faster than the big names? That’s no small feat. Their focus on reasoning models is especially interesting and definitely shaking things up in the AI space. #AI #DeepLearning #Innovation #TechTrends #MachineLearning #Startups https://lnkd.in/dDfmM99D

6

Just open-sourced my real-time GE flipper for Old School RuneScape (OSRS) – built with Next.js 16, React 19, and Zustand. It's a portfolio piece that's actually useful for spotting flips in the Grand Exchange grind. Handles 30k+ items with live price updates every 5 minutes and custom expression columns for advanced filtering. ➡️ Real-time Data Handling: Pulls and processes massive OSRS item datasets from the API, updating prices live to help identify profitable flips without manual refreshes. ➡️ Customizable UI: Built with React 19 hooks and Zustand for state management – add your own calculated columns (e.g., profit margins, ROI) via simple expressions. ➡️ Performance Optimized: Next.js 16 for server-side rendering and static generation, ensuring snappy load times even with thousands of items in view. Check it out on GitHub: https://lnkd.in/gK6BzMXD (Yes, I still play OSRS. Remote dev life keeps the nostalgia alive.) #OSRS #OpenSource #NextJS #React #WebDev #RuneScape

14

If you're keen on real AI use cases, this is actually quite powerful and I encourage you to read further (and help us build: https://lnkd.in/gsYnAfKK) At Hex, we've found our Threads product (https://lnkd.in/gJKCurJB) is the best way to summarize customer support load, because we're not limited to what's in the customer support portal - we can ask in terms of active users, customer size, feature adoption, and still get great answers. This is based on a strategy of keeping your data warehouse as a source of truth, and all related data work in one place - Hex - so that AI understands concepts in your business and makes smart connections between things. Arguably, ALL data analysis for a particular vertical is eventually best achieved in a tool like Hex, connected to curated data in your data warehouse. Wait - can't I do this directly in the customer support tool? Can't I connect an MCP server and get a functionality like this in one of my other AI enabled products? Maybe, but you have 100+ SaaS apps, so play this scenario out: - How many places are you going to sync your core data sets? - How many cross product integrations are going to exist, and how many of them are you going to enable? (100 choose 2 is ~5000) - In how many of these products will all users have seats? - In how many of these products will these seats have permissions to use AI features? (Our support tool just told me I don't have permission to ask the agent questions, probably because I don't have the right seat type, which means more $$) My bet is a on a world with ONE place for all analysis.

33

2 Comments

A little while ago I posted about Sacramento Labs being interested in acquiring software products and heard from a number of folks I hadn’t talked to in a while. The responses and conversations were genuinely great. (Original post here: https://lnkd.in/g-6hjy6q) After talking it through, Garrett Milster and I decided to formalize something we were already doing informally: we’re offering a $10,000 referral fee for any warm introduction to a company that we end up acquiring. A few notes for clarity: -The introduction needs to be a direct, warm introduction (email or message). -Public listings or cold forwards don’t qualify. -Referral fees are paid after closing and are contingent on a completed acquisition. -There’s no expectation of ongoing involvement unless explicitly discussed. If you come across a software business that feels like it might be a fit for us, feel free to make an intro.

6

We just shipped 𝗦𝘆𝗻𝘁𝗵𝗲𝘁𝗶𝗰 𝗨𝘀𝗲𝗿𝘀 on Reforge Build. AI agents that open a real browser, take on distinct user personas, and evaluate your product end to end. They navigate flows, read content, interact with features, and form real opinions about the experience. The feedback isn’t surface-level QA, it’s product-level. Messaging clarity. Value prop gaps. What would make someone convert or bounce. You set up personas and the specific tasks or flows you want tested. Every combination runs in parallel. What used to take weeks of recruiting and interviews comes back in minutes. What I’m most excited about is the scope we’ve unlocked for Build. Synthetic Users works on any public URL, including prototypes, staging previews, and live products. Product discovery doesn’t start and stop before launch. Understanding how different user types experience what you’ve already built is half the battle, and most teams don’t do it enough because the traditional approach doesn’t scale. 🧤 Try it here: https://lnkd.in/gZdMDwSX — The most fun part of bringing this product to life was the AI engineering work around the agents’ behavior. Computer use agents are great at finding a piece of information on a page. They love filing bug reports about a border radius being inconsistent. Getting browser agents to look beyond surface-level feedback and instead be able to say: “𝘌𝘢𝘤𝘩 𝘵𝘪𝘦𝘳 𝘭𝘪𝘴𝘵𝘴 𝘢𝘥𝘥𝘪𝘵𝘪𝘰𝘯𝘢𝘭 𝘤𝘢𝘱𝘢𝘣𝘪𝘭𝘪𝘵𝘪𝘦𝘴 𝘢𝘯𝘥 𝘭𝘢𝘳𝘨𝘦𝘳 𝘶𝘴𝘢𝘨𝘦 𝘭𝘪𝘮𝘪𝘵𝘴, 𝘣𝘶𝘵 𝘪𝘵’𝘴 𝘯𝘰𝘵 𝘤𝘭𝘦𝘢𝘳 𝘸𝘩𝘪𝘤𝘩 𝘰𝘶𝘵𝘤𝘰𝘮𝘦𝘴—𝘦.𝘨., 𝘦𝘯𝘥-𝘵𝘰-𝘦𝘯𝘥 𝘫𝘰𝘶𝘳𝘯𝘦𝘺 𝘮𝘢𝘱𝘱𝘪𝘯𝘨 𝘷𝘴. 𝘣𝘳𝘰𝘢𝘥𝘦𝘳 𝘤𝘰𝘩𝘰𝘳𝘵 𝘢𝘯𝘢𝘭𝘺𝘴𝘪𝘴—𝘢𝘳𝘦 𝘶𝘯𝘭𝘰𝘤𝘬𝘦𝘥 𝘢𝘵 𝘦𝘢𝘤𝘩 𝘱𝘳𝘪𝘤𝘦 𝘱𝘰𝘪𝘯𝘵.” …forced us to rethink how long-running agents prioritize signals and form judgments. But that’s the feedback that actually moves the product forward. Huge thank you to teammates at Reforge, especially Stewart Eaton and Chun Jiang, for helping bring these agents (users? friends?) to life. And a big shoutout to the Browserbase team for engineering such a great product to build on (cc Arik Bird).

60

15 Comments

Sarah Guo's piece on software abundance nails something important: when code generation gets commoditized, the bottleneck shifts. Not to writing code. To specifying intent, surfacing hidden assumptions, exercising judgment. But there's a piece that most of the abundance discourse misses. You can articulate intent perfectly and still ship something broken, because the agent didn't know what was actually running in production. It didn't know about the job that's been silently failing. It didn't see the edge case baked into the pipeline three months ago by someone who's no longer on the team. Organizational knowledge doesn't live in your prompt. It lives in your production system. The real unlock isn't just better intent specification. It's closing the loop from production back into the development cycle, continuously, so agents are operating on reality, not assumptions. That's how software is built - in iterations and reaction to feedback, until we get it right. That's why the agile methodology was such a game changer. That's what software abundance actually requires. Want to learn more about closing the production feedback loop? check it out: https://lnkd.in/dx9pYrPq

49

4 Comments

I'm helping a few friends find technical co-founders, early-stage CTOs, and heads of engineering. These are exited founders and people with real traction and backing from top VCs.  If you are a ridiculously talented engineer, you owe it to yourself to take a big swing in this generational paradigm shift. I can't guarantee success, but I can promise long hours, high stress, meaningful work, ownership, agency and insane growth. Which is to say, it'll be a hell of a ride. US only for now. If this is interesting, hit me up. Note: I'm not a recruiter, I'm truly just helping some people I think are talented.

61

Our RAG paper hit front page The Hacker News last year. A VC just asked me if we still use Pinecone. Here's why everyone who mastered RAG might need to start over: We spent months perfecting Reciprocal Rank Fusion for hybrid search. The problem was clear: Vector-only search failed on specific queries. When users asked: "What features are included in a premium plan?" Vector search would pull marketing materials and testimonials. But it would miss the actual premium plan documentation. So we built a hybrid approach: - Vector search for semantic matching - Keyword search for exact term targeting - Reciprocal Rank Fusion to combine results The results were impressive. - Better accuracy across our diverse customer base. - Minimal parameter tuning required. - Robust performance without overfitting. We integrated Pinecone for vectors and Algolia for keywords. Avoided the maintenance burden of self-hosted solutions. Focused our engineering resources on core AI functionality. But here's what happened next: Our internal evaluations show agentic search outperforming RAG entirely. Instead of vectorizing documents and pulling top-N results, the new approach fans out across sources using reasoning models. - It evaluates multiple sources. - Composes answers dynamically. - Delivers higher quality than traditional retrieval. The biggest gains in cost and performance came from entirely rethinking the search stack. Not from reranking or query recomputation. Overall, we might rip out a lot of our RAG tech stack. The retrieval approach already feels outdated and agentic search is already outperforming it. And the speed of change means you can spend months perfecting something, only for it to be obsolete a few months later. That's how fast this space moves. Everyone spent months getting good at vector databases and fusion and that expertise might already be obsolete… unfortunately. The lesson is this: Stay on the cutting edge because the ground will shift beneath you every six months or so. What technical skills are you betting will still matter next year?

36

3 Comments

I was chatting with another founder in the AI space recently, and we both think that the market for junior engineers is going to flip. Here's why: I keep seeing links to graphs showing that senior engineers are being hired more often, and junior engineers are being hired less. I think this is going to change due to "speed of shipping". We've put new grads on projects who have spent the last four years using claude and chat gpt to do homework, write resumes, and get through classes. They absolutely CREAM senior engineers on shipping velocity. Eager 22 year olds who want to make an impact now can lean on helpers like claude code to make up for experience. Not only that, but folks in the same cohort as I am are slower and more methodical, even with those tools. The perfect blend? Someone with experience and taste reviewing PRs, writing out architecture (because coding agents aren't as good at that) and holding the reins, PLUS a few junior engineers is the perfect mix. It's the team of the future. Do you agree? Disagree?

16

9 Comments