Shannon Tan

Curious: has anyone found a particular sub-agent role or setup that's been especially powerful? Would love to compare notes. I've been experimenting with Claude agent teams lately, and one thing stands out: the software paradigms we've always relied on (API design, test coverage, modularity) matter more when working with AI, not less. Giving each sub-agent a specific role (Team Lead, Frontend Dev, Backend Dev, QA Engineer) produced noticeably better results than asking one agent to do everything. Wrote up what I learned using a calculator app as the example. Link in the comments.

12

1 Comment

Hiring engineers for hard operational problems is different. Most candidates can write clean code. Fewer can sit with a dispatcher at 7am and understand why their shift planning breaks down. Fewer still can go back to the codebase and design something that actually works in production with those constraints. What I look for at Mirage Metrics is not just skill with Python, SQL, or distributed systems. It is a willingness to go into environments that are messy, sometimes chaotic, and keep digging until the real bottleneck shows up. That can mean inventory mismatches hidden in Excel, planning rules written on a whiteboard, or API endpoints that return inconsistent data depending on who queries them. The best engineers I have worked with are the ones who do not flinch at this. They treat debugging a customs declaration pipeline with the same seriousness as debugging a memory leak. They know that solving “boring” problems in data quality or workflow mapping is what unlocks everything else. On the technical side, I test for clarity. Can someone explain the trade-offs between running a model via an API versus hosting it on GPUs without getting lost in jargon. Can they design a schema validator that will still make sense six months later when requirements change. Can they write glue code that is robust, not fragile. On the personal side, I look for stamina. These projects are rarely about building a shiny feature in isolation. They require iteration with operators, late-night debugging of OCR failures, and the patience to integrate into legacy systems that were never designed for AI. If you want to work on AI for logistics, manufacturing, or mining, the question is not only whether you can code. It is whether you can hold your ground in the real world, in front of the people whose work depends on your system. That is the bar.

17

5 Comments

I've sat in on 120+ unique DevTool sales calls in 2025. Every company that was able to book meetings with engineering leaders had these 3 strategies in common: 1) They Fish Where the Fish Are Their sales reps get dynamic target account lists on a weekly basis based on actual developer activity signals. Not quarterly updates from last year's master ICP account list. These accounts are already in-market, showing intent through their engineering team's behavior. Without this, you're basically handing your SDRs a list of companies to annoy. 2) The Multi-Threading Strategy Here's the reality: CTOs/Engineering leaders with budgets are often detached from the ground-level product evaluation happening in their teams. Your tool might be battle-tested by 15 engineers while the VP has no clue it exists. Smart companies run a two-pronged approach: • Sales team targets economic buyers (top-down) • Marketing & DevRel nurtures the developers (bottom-up) In an ideal scenario: When your SDR finally gets that VP on a call, half the work is already done because the engineering team is already advocating internally or deeply invested in building using your solution. 3) They Have Real Product-Market Fit Look, I've been on both sides of this. I've sold products that had zero PMF and products that people actually wanted. The difference is massive. Even if you nail the timing, targeting, and messaging—everything falls flat if you're not solving a genuine pain point that prospects will pay to fix. In DevTools, this matters even more because developers talk. A lot. Great tools spread through the community faster than drama in a startup Slack channel. Heck, Reo.Dev is barely 2 years old, but we've grown entirely through cold LinkedIn/email outbound because our messaging hits real pain points our prospects face daily. And we are not even a DevTool! So many DevTool founders & Sales head keep asking me: "How did you crack the LinkedIn outbound game? What playbooks are your most successful customers using?" After getting this question for the 100th time, my team and I spent the past month condensing everything we know about DevTool LinkedIn outreach into one comprehensive webinar. The perk of my job? Front-row access to 120+ developer-focused companies and their GTM strategies. I've seen what works, what fails spectacularly, and the tactics that create outlier LinkedIn success. 📆 We're sharing it all—for free! Live—on August 27th at 1 PM EDT. Trust me, you don't want to miss this one. You won't find this content anywhere on the internet. That is my promise to you, and if you show up live you will know why :) 🔗 Link to register is in comments.

30

3 Comments

You might have a functional website, but that doesn't mean it works. Today I learned about a unique California alert system and went to the CHP website to discover more about it. Only I didn't Instead I got a big old no response in my browser. Confused I double checked the availability of the website and quickly confirmed that there was something there that triggered my DNS level ad filtering. This isn't surprising as I maintain a lot of rules and prefer more blocking to less, but I was certain this had to be a mistake likely triggered by a bad filter. Turns out I was wrong, the filter was in fact working correctly, only the CHP seems to have chosen a specific hoster that has a reputation for being a source of trackers. Feel how you will about their choice of hoster, but there's an interesting problem here. What's the right way to solve picking a bad host? While lifting and migrating to a new hoster is definitely an option, what's the likelihood it will happen? Until then the CHP website is down for anyone who uses the Peter Lowe adlist and if they are non technical there's essentially no solution for them. Problems like this aren't going to go away and everyone should be taking notice. California is in a very good spot and they should be taking note. Ironically California has already made some AMAZING plays that lay the ground work for this. Californians stop screaming, this might be one of the only compliments ever given to your state government. In 2023 California announced their intention to partner with a generic medicine manufacturer to produce insulin for 10 years. 3 years later they were shipping insulin to the market with copays 1/10th of the federally maintained copay. The program has been a resounding success. How can I say this so surely? I'm a type 1 diabetic and the production of Californian insulin has created one of the only markets in the USA where a Type 1 can actually have a hope at the pharmacy. California is also the only state in the Union now that has a chance of negotiating with ANY insulin producer (it's not easy to negotiate when playing hardball kills the people you are fighting for). The model from 2023 should be at play in states like California (or Washington), or any state looking to gain an economic advantage into the future. States need to build their own cloud competitor.

Just wrapped up a tricky server-side migration for our data ingestion pipeline, specifically the SFTP piece. The goal was simple: move everything over without clients noticing anything had changed. The initial design was a bit more involved. We planned to write an onboarding script that would mirror our current process, but only for SFTP. Key Vault secrets would be copied over, and a Python script in our ADO pipelines would recreate the user accounts. We even added verification stages to ensure the users matched on both servers. In theory, it was solid. In reality, it broke down fast, too many components, too fragile, and we were running up against a deadline. So I took a step back and proposed a simpler approach: write a focused C# script to back up the entire SFTP server and restore it to the new one. Mirror directories by walking through the existing structure, replicate Key Vault secrets directly. One pipeline, no complex branching logic, just a straightforward backup and restore. It worked. The migration was successful, and we avoided introducing extra complexity into something that didn’t need it. Sometimes, the best solutions come from leaning into simplicity.

2

We asked more than 600 engineering professionals what they think about AI, investment, metrics and so much more for this year’s State of Engineering Management report. Here’s sneak peek of what we found 👀 🛠 48% use AI to write code ⚡ 62% report 25%+ productivity increases 💬 81% believe a quarter of today’s engineering work will shift to AI within 5 years You can read the full report here → https://lnkd.in/eD3899_B

5

SWEs have LeetCode. Quants have QuantGuide. PMs? They now have the PMGrind.com, offering all-in-one PM prep for 24/7 practice. Two months ago, I made a post about the pains of PM recruiting that went viral. It got~1M hits, sparked 4.5K+ comments, 1K+ reactions, and filled my inbox with hundreds of DMs (and counting!). The message clearly struck a nerve: PM recruiting is broken. 🛠️ Aspiring PMs are stuck: ❌ Wasting 2 hours for every 1 hour of mock practice due to mandatory role-switching with peers. ❌ Receiving inconsistent feedback from other inexperienced candidates. ❌ Struggling to practice on their own schedule. So Aneesh Patnaik and I founded the PMGrind.com: ✅ AI-powered mock interviewing agent — simulates real PM interviews with dynamic back-and-forth (no clunky “press enter to continue”). ✅ Scoring engine trained on top company rubrics like Google, Amazon, and Meta — custom-tailored to your question type. ✅ Free — no waitlists, no awkward favors. 🚀 To celebrate the launch, I’m doing free resume reviews! Comment what company you want to see next on the PMGrind.com, and you’ll be entered to win a free review from me! 📄🔍 #productmanagement #pm #recruiting #leetcode #ai #agent #mock #interview

189

18 Comments

This week's blogpost on the holy shit gap [0] got a bit more traction than expected so here's a promised follow up on something that made me, personally, go "holy shit I cannot believe this worked" For some learning purposes (and some shits and giggles) I've been trying to learn more about how Claude Code works under the hood, and write a proxy so I can proxy requests to some models I control (with a non-compatible API). I had jury rigged a proxy (by inspecting network traces) and it was working *ok* but not great - there were subtle bugs. I found myself saying "hmm I wish I knew what claude really was doing with xyz" ... and then I realized -- hey... I can do this. So I asked claude code to decompile itself, figure out where network requests were happening, explain the architecture (which alone fixed a few bugs), and then add some logs and tracing and then I had a loop where I said “Hey Claude, a Databricks engineer reported this, could you add logs to claude code so we can fix/investigate” and then “Hey Claude, an anthropic engineer gave this response, can you fix this” in a loop and it ironed out a number of bugs, made the implementation much more spec compatible and more robust. Sure, it took a few hours on a friday afternoon and cost about 60 million sonnet tokens, but hey, the thing *worked* and unblocked me. And I still can't believe it. Attached is a very early screenshot of some of the responses, before things got way too detailed and specific to my logs.

66

1 Comment

For my remote work friends: Would you be interested in a spreadsheet of remote-work-friendly locations (WiFi speed, noise levels, etc)? I've cataloged ~30 cafes to date in/around the greater Seattle area, and I find it's very useful for deciding where to meet up with new folk to work with. If I may share my top 5: 1: A Muddy Cup (Wallingford) - 9.5 2: Boon Boona (Renton) - 9 3: Patcha Collective (Greenlake) - 8.5 4: SBP West Wall (Atlantic) - 8 5: Cafe Jaiiya (Edmonds) - 8

15

4 Comments

We put every engineer in every customer Slack channel. No middle layers. No ticketing system buffer. Just direct connection. They’re the ones replying to questions, squashing bugs, and turning feature requests into real updates. It’s the fastest way to feel customer pain, and the fastest way to build empathy. The result? • Engineers feel the real urgency of customer problems. • Customers feel like they’re talking to the people who actually make the product. • The feedback loop is lightning fast. Curious how others do this. Do your engineers talk directly with customers, or is that still a rare thing?

118

62 Comments