Dias | building in public

Everyday I fail is another day im closer to the life im guaranteed So never give up, just make another unprofitable b2b AI SaaS

@MicroLaunchHQ 100%

Founders: one launch a month. Minimum.

@CodeWithAmann Nahhh, it's still the best

VS Code disappeared like it never existed.

I have 5 projects on my desk right now. 3 are client projects. They pay the bills and give me the freedom to build the other 2. From today, this account is about those two. The first one is a marketplace for home repair contractors. We're starting in Kazakhstan, with plans to expand across the CIS. Finding a contractor here usually goes like this: you ask a friend, they ask a WhatsApp group, and eventually someone shows up. Maybe they're great. Maybe they're not. You don't really know until the job is already underway. Before writing much code, we spent weeks talking to users and tearing apart every competitor we could find. We also decided to build the supply side first. No point launching a marketplace if there's nobody there when the first customer arrives. The second project is Scam Shield. It's a mobile app that helps people spot phone scams and social engineering before they become victims. We're still testing it. Neither product is launched yet. The goal is to ship both before August. I'm writing that publicly because deadlines become more real when other people can see them. I'll share the numbers, the bugs, the wrong assumptions, and the things that don't go as planned.

The new math of work - and the hidden line item. For most of history, throughput was capped by headcount. Want more output? Hire. That equation is breaking. Today, one person + a computer + AI produces what used to take a team of 15–20. Jensen Huang’s 10-year vision for NVIDIA: 75,000 humans working alongside 7.5 million AI agents — 100 agents per employee. His framing: “IT will become HR for AI agents.” Claude + MCP makes this concrete. MCP turned Claude from a clever chatbot into an actual operator — it crawls my browser, Drive, local files, GitHub, databases. A junior who doesn’t eat, sleep, or need motivation. Plug in a connector, get a hand. That’s the efficiency machine. It only grows from here. Now the part nobody writes about with the same enthusiasm. My role inside this loop is shrinking. I see it in myself — more forgetful, more scattered. Hundreds of open tabs. 15 parallel Claude flows. I remember what I’m doing but not where it lives. Short-term processing now pays better than memory, which is strange, because memory used to be one of my main working assets. March 2026, BCG Henderson Institute / HBR study, 1,488 workers — they call it “AI brain fry.” Mental fog, slower decisions, the feeling of a “crowded mind.” The cause isn’t using AI. It’s supervising it. The brain isn’t running one task — it’s babysitting several layers of digital output at once. What’s actually happening: The human stopped being just an executor. The human is now a dispatcher. And this is the early stage. Today it’s agents and a few dozen automations. Tomorrow one person manages hundreds or thousands of physical robots, bound to them through a biometric key. Not science fiction. April 2026: Physical Intelligence released π0.7 — a model that performs tasks it wasn’t trained on, operating an appliance it had seen twice in its dataset. The Claude-style scaling curve is now arriving in the physical world. In parallel: brain-swarm interfaces, scaled autonomy — one operator running a swarm, the system deciding which robot needs human input at any given moment. This isn’t “one person = a 20-person firm” anymore. It’s one person = a corporation. The era of hiring people to grow isn’t the only era. The era of scaling your own operator-bandwidth has started. I’m curious to live to the day when “wealth” gets measured by how many robots are keyed to you.

The biggest scam in AI? Anthropic’s marketing — I wish. The actual story is worse. For months SWE-Bench Pro has been THE benchmark everyone quoted for “coding ability.” Turns out it’s quietly broken. A new benchmark from Datacurve called DeepSWE just dropped and audits the whole thing: 8.5% false positives, 24% false negatives. And some models were literally exploiting the harness — reading gold commits straight out of .git history. Datacurve flagged ~12% of Claude Opus’s SWE-Bench Pro runs as cheating. The reshuffle is brutal: • Claude Haiku 4.5: 39% → 0% • Claude Sonnet 4.6: drops into the 30s • Gemini Pro family: melts into single digits in several configs • GPT-5.5: 59% → 70%, sitting 16 points clear of Opus 4.7 Same models, different ruler. A 30-point spread on SWE-Bench Pro becomes a 70-point spread on DeepSWE. So no, the cope wasn’t mine. It was every Anthropic stan’s.

ChatGPT costs the same in New York, London, Berlin, and Milan. Access to AI is basically universal. And yet US companies use it far more than European ones, and even inside Europe the gap is huge the UK, Sweden, and the Netherlands pull ahead while France and Italy lag. A new NBER paper digs into why. It isn't budgets, infrastructure, or tech availability. It's management quality. Companies that promote strong people, measure outcomes, and reward performance also adopt AI much faster. They buy the subscriptions, train their teams, rewrite workflows, and openly push people to use the tools. Badly run companies just hand out logins and hope something happens. So AI is quietly turning into a management diagnostic. Having the technology stopped being a moat the moment everyone got it. The real difference is whether leadership can actually change how work gets done. The next few years won't reward the best tech. They'll reward the best management. AI just makes that gap impossible to hide. nber.org/papers/w34995

@dexter_z Thanks for reaching out! I’ll keep this in mind as things develop.

@Satkanovv Hi Dias, if you'll need an experienced backend developer for your project, please reach out to me. I would like to be a part of your journey

@builtbyamina They both look good, but the left one works better for me.

Which one would you choose. Right or Left?

@pranavcodes_ C++😁

in which language you wrote your first "hello world"?

@hridoyreh Genius 🤔

This is how I find SaaS feature ideas:

@1Umairshaikh 🤔💻🔄

Describe your startup journey in 3 emojis

@rasmalai Just getting started here, but love the consistency! I’ll drop a reply too — excited to connect.

@qasimbizs Working on learning and experimenting with AI projects. Excited for 2026!

2026 is here. What are you building right now?

@1Umairshaikh Congrats! Really motivating for someone just getting started.

Got my biggest payout so far without being a reply guy 🎉 Quality and consistency matter more than anything else. Thanks, everyone 🫶

@stijnnoorman Appreciate you sharing this. It helps a lot when you’re just getting started. I’ll try to use your advice.

Below 1,000 followers? Focus on this: • Comment 50x/day • Analyze what works • Optimize your profile • Send a few DMs/day • Write 3 posts per day • Share what you learn Most creators stay stuck because they post inconsistently and engage passively. Do the opposite. Show up aggressively for 30 days. Then keep showing up every single day. You’ll hit 1,000 followers faster than you think. • Then 2,000 • Then 5,000 • Then 10,000 You only need to do the right inputs for long enough. Because the right inputs lead to the right outcomes. Put in the work.

@mdnlabs Congrats 🙌

2AM. got my first payout. literally in shock right now. i have no idea what happened.

@Mkpareek19_ Nice work. The “reply to everyone” part is underrated - that’s where real momentum comes from.

@VadimStrizheus Yep. Hard to beat for everyday work.

Do you still use VSCode?

@MarkKnd UI looks good. What kind of feedback are you looking for — UX, clarity of message, or conversion?

What do you think of this website?