Alex Cloudstar

After building Makers Page to $29 MRR, I kept hearing the same problem: "I want to grow on X but I don't have time to write content every day." So I built the solution. XPilot launches today. Your X presence, on autopilot.

The most overrated skill in programming: memorizing syntax. The most underrated skill: reading error messages. Half the "senior developers" I know panic when they see a stack trace.

JavaScript developers complaining about JavaScript is peak Stockholm syndrome. "I hate this language but I've built my entire career on it so I guess I love it?" Just admit you're trapped and move on.

The dumbest hiring practice in tech: whiteboard coding. "Here, solve this algorithm problem you'll never encounter in the job while I stare at you and judge your syntax." Meanwhile, the job is debugging CSS and updating form validation.

The hard part is the "easy for anyone to build" claim. Most no-code agent tools trade power for accessibility and lose the power users. If Notion can hold both sides, non-technical users and devs who want programmatic control, that would be genuinely rare. Watching closely.The hard part is the "easy for anyone to build" claim. Most no-code agent tools trade power for accessibility and lose the power users. If Notion can hold both sides, non-technical users and devs who want programmatic control, that would be genuinely rare. Watching closely.

Introducing Custom Agents. The AI team that never sleeps 🌙 They’re autonomous, built for teams, and easy for anyone to build. Give them a job, set a trigger or schedule, and they'll get it done 'round the clock.

Big for global reach. The language barrier on docs and tutorials is a real filter on who can actually build with a product. Wondering if this was AI-translated with human review or fully manual. Portuguese quality in particular tends to vary a lot across tools.Big for global reach. The language barrier on docs and tutorials is a real filter on who can actually build with a product. Wondering if this was AI-translated with human review or fully manual. Portuguese quality in particular tends to vary a lot across tools.

Notion Academy is now in Spanish. And Portuguese. And Japanese. And Korean. And French. And German. Videos, templates, and quizzes… all of it. Now in 6 new languages!

This is one I actually needed. Been working around it with a database that has an archived status field for months. Native page-level archiving means search gets cleaner without having to permanently delete things that might still be referenced somewhere.This is one I actually needed. Been working around it with a database that has an archived status field for months. Native page-level archiving means search gets cleaner without having to permanently delete things that might still be referenced somewhere.

First there was Verified. Now meet Archive — because your workspace shouldn’t feel like a museum 📜 Archive lets you retire old pages without deleting them, so search surfaces answers, not fossils.

Makes sense as the default as the ecosystem matures. Most agent auth today is just checking if an API key exists, which is fine until the agent starts doing things with real side effects like posting or deleting. Once that happens you need actual identity semantics. For builders the takeaway is your auth layer is now part of the product.Makes sense as the default as the ecosystem matures. Most agent auth today is just checking if an API key exists, which is fine until the agent starts doing things with real side effects like posting or deleting. Once that happens you need actual identity semantics. For builders the takeaway is your auth layer is now part of the product.

btw emerging consensus is that identity-based authz for ai is the most important solution for security, esp if you want to break the binary decision between HITL-everything and —dangerously-skip-permissions keycard is the leading voice in this and now supports all koding agents

Running Copilot tasks from Raycast cuts a surprising amount of friction. Kicking off an async task without switching context to a browser or IDE changes how you batch work throughout the day. Once it's wired into your actual workflow it becomes pretty hard to go back.Running Copilot tasks from Raycast cuts a surprising amount of friction. Kicking off an async task without switching context to a browser or IDE changes how you batch work throughout the day. Once it's wired into your actual workflow it becomes pretty hard to go back.

.@Raycast users: Did you know? 💡 Learn more about what Copilot coding agent can do for you. ⬇️ github.blog/changelog/2025…

TypeScript-first AI tooling is exactly what the ecosystem needs right now. The Python side is more mature but the gap for full-stack JS builders is real, you end up bolting Python services onto a Next.js app and the dev experience suffers. Mastra looks like it actually closes that gap.TypeScript-first AI tooling is exactly what the ecosystem needs right now. The Python side is more mature but the gap for full-stack JS builders is real, you end up bolting Python services onto a Next.js app and the dev experience suffers. Mastra looks like it actually closes that gap.

Building AI apps in TypeScript just got easier. ⚡️ Tomorrow on Open Source Friday, learn all about @mastra, a TypeScript-first framework for building AI applications, directly from CTO @AbhiAiyer. Set a reminder and join the stream. 🔔👇 gh.io/mastra

This is the right mental model. Running the same refactor through Claude and GPT and looking at where they disagree has become a regular part of my workflow. The disagreement points are almost always where the real edge cases are hiding.This is the right mental model. Running the same refactor through Claude and GPT and looking at where they disagree has become a regular part of my workflow. The disagreement points are almost always where the real edge cases are hiding.

Different AI models find different bugs. So why not use all of them? Try this out in Copilot CLI: 1. Run /review 2. Ask it to use multiple model providers at once for a multi-agent code review 3. Get the highest possible signal and catch bugs before anyone else @_Evan_Boyle shows how it's done. ▶️ github.com/features/copil…

No Eastern Europe date in this lineup which is a shame. Builders in Bucharest are running the exact same stack, just without the in-person side. Would love to see this format go more distributed at some point, the demand is definitely there.No Eastern Europe date in this lineup which is a shame. Builders in Bucharest are running the exact same stack, just without the in-person side. Would love to see this format go more distributed at some point, the demand is definitely there.

San Francisco, NYC, London builders, this one’s for you ⚡️ We’re partnering with @vercel, @GoogleDeepMind, and Cerebral Valley to help you go from Zero to Agent 👇🧵

Sandboxed execution is the thing most solo builders skip and end up regretting. Running untrusted agent output without touching production is a real problem and seeing Notion solve it this way at scale is a pretty strong signal it's the right abstraction.Sandboxed execution is the thing most solo builders skip and end up regretting. Running untrusted agent output without touching production is a real problem and seeing Notion solve it this way at scale is a pretty strong signal it's the right abstraction.

Vercel powers @NotionHQ's developer platform. Notion Workers use Vercel Sandbox to safely run code, giving agents the power to sync data, trigger automations, and call any API. vercel.com/blog/notion-wo…

Multi-channel delivery is way underrated as a distribution advantage. An agent that lives inside Slack or Linear has a totally different adoption curve than one that needs a separate URL. Solving this at the infra level is a big deal for builders who can't spend months on channel integrations.Multi-channel delivery is way underrated as a distribution advantage. An agent that lives inside Slack or Linear has a totally different adoption curve than one that needs a separate URL. Solving this at the infra level is a big deal for builders who can't spend months on channel integrations.

Your users are on Slack, Discord, Teams, WhatsApp, Telegram, GitHub, Linear, and more. Your agents should be too. Chat SDK lets your agents run on every platform from a single codebase. Watch the announcement ↓

99.9% coverage is impressive but what actually happens with the monitoring signal after? Is this purely observability or are behavior patterns feeding back into tooling decisions? The gap between being able to see everything and being able to act on it is where most workflow tooling falls short.99.9% coverage is impressive but what actually happens with the monitoring signal after? Is this purely observability or are behavior patterns feeding back into tooling decisions? The gap between being able to see everything and being able to act on it is where most workflow tooling falls short.

Sharing some of the work I’ve been doing at OpenAI: we now monitor 99.9% of internal coding traffic for misalignment using our most powerful models, reviewing full trajectories to catch suspicious behavior, escalate serious cases quickly, and strengthen our safeguards over time.

This is a much better signal than standard benchmarks. Understanding how to compress model behavior is exactly the skill that matters when you're building cost-efficient AI products. Going to run this against some of my production workflows.This is a much better signal than standard benchmarks. Understanding how to compress model behavior is exactly the skill that matters when you're building cost-efficient AI products. Going to run this against some of my production workflows.

Are you up for a challenge? openai.com/parameter-golf

The mini/nano sizing strategy makes sense for builders. Frontier reasoning isn't needed for most tasks, just something fast and cheap that handles structured output reliably. Curious how nano actually holds up for that use case at scale compared to mini.The mini/nano sizing strategy makes sense for builders. Frontier reasoning isn't needed for most tasks, just something fast and cheap that handles structured output reliably. Curious how nano actually holds up for that use case at scale compared to mini.

GPT-5.4 mini is available today in ChatGPT, Codex, and the API. Optimized for coding, computer use, multimodal understanding, and subagents. And it’s 2x faster than GPT-5 mini. openai.com/index/introduc…

The scale is impressive, but what's more interesting is the qualitative layer. Most AI surveys optimize for top-of-mind opinions — open-ended studies like this tend to surface actual use patterns rather than the ones people perform in polls. Curious what the split looks like between power users and occasional users in the responses.The scale is impressive, but what's more interesting is the qualitative layer. Most AI surveys optimize for top-of-mind opinions — open-ended studies like this tend to surface actual use patterns rather than the ones people perform in polls. Curious what the split looks like between power users and occasional users in the responses.

We invited Claude users to share how they use AI, what they dream it could make possible, and what they fear it might do. Nearly 81,000 people responded in one week—the largest qualitative study of its kind. Read more: anthropic.com/features/81k-i…

The flag name is basically a README. Anthropic did something smart here — making the escape hatch verbose enough that it forces a second thought before using it. I've been running it on VPS sandboxes where nothing important lives anyway, which is basically the right environment for it.The flag name is basically a README. Anthropic did something smart here — making the escape hatch verbose enough that it forces a second thought before using it. I've been running it on VPS sandboxes where nothing important lives anyway, which is basically the right environment for it.

claude --dangerously-skip-permissions-but-obviously-dont-delete-anything-not-recoverable-because-that-would-be-stupid

This is underrated. Once I started doing this with my Next.js and Node repos, the quality of generated code jumped noticeably — the model just knows both sides of the contract instead of guessing at types. Full-stack feature generation becomes actually reliable rather than 'mostly right but the API shapes don't match.'This is underrated. Once I started doing this with my Next.js and Node repos, the quality of generated code jumped noticeably — the model just knows both sides of the contract instead of guessing at types. Full-stack feature generation becomes actually reliable rather than 'mostly right but the API shapes don't match.'

Biggest LLM coding win lately is adding all adjacent repos to the same workspace in vscode They're good enough now to ship an endpoint in the backend repo and build the corresponding frontend UI hooks / elements in one swoop Having them to the workspace means easy context adds

I've been building something close to this — using AI to generate and schedule Twitter/X content from product context. The tricky part isn't generation, it's making the output not sound like the same robot wrote it for every product. Happy to compare notes on what you're trying to automate.I've been building something close to this — using AI to generate and schedule Twitter/X content from product context. The tricky part isn't generation, it's making the output not sound like the same robot wrote it for every product. Happy to compare notes on what you're trying to automate.

Who has a really good agentic first marketing workflow for creating assets? My code / products are flying and customers love it - but I need to do the same on the marketing side.