Sandeep (@sandeepg33k) 's Twitter Profile
Sandeep

@sandeepg33k

co-founder @hashnode. Still writing your docs and blogs with stone-age tools? 🛠️ Upgrade to Hashnode and unify your API docs and blog like it’s 3024!

ID: 92261165

linkhttps://hashnode.com calendar_today24-11-2009 12:30:21

6,6K Tweet

6,6K Takipçi

456 Takip Edilen

Sandeep (@sandeepg33k) 's Twitter Profile Photo

Just integrated Vercel AI Gateway at Bug0. I think it’s pretty great for a couple of reasons: - no extra markup - you can BYOK - in case of failures, they automatically retry using their system creds so you get extra reliability and much higher rate limits - send

Sandeep (@sandeepg33k) 's Twitter Profile Photo

Always bet on LLMs to get better. When we started Bug0 ~3 months ago, I wasn’t sure if letting AI directly test web apps was the right decision. To get determinism, we started by generating stable locators and building AI-powered auto-healing. Fast forward to today: we’ve

Always bet on LLMs to get better.

When we started <a href="/bug0inc/">Bug0</a> ~3 months ago, I wasn’t sure if letting AI directly test web apps was the right decision. To get determinism, we started by generating stable locators and building AI-powered auto-healing.

Fast forward to today: we’ve
Sandeep (@sandeepg33k) 's Twitter Profile Photo

We just crossed 10K test executions at Bug0 in the last 3 months. My biggest learning in getting LLMs to do what you want: - LLMs are non-deterministic tools. Do not over-optimize trying to make them deterministic. - use tool calling to do tasks deterministically. - use

DX Tips Magazine (@dxtipshq) 's Twitter Profile Photo

🆕 Every week, Michelle Fang 🌁 tracks dozens of developer meetups in SF. There was a Thursday last year with 10 (!) AI meetups. It's really hard to stand out. dx.tips/hosting Sarah Chieng guests with how she gets over 1000+ RSVPs and way more online) for her events:

Dane Knecht 🦭 (@dok2001) 's Twitter Profile Photo

If one of your angel investors berates you into using their product, they aren't going to be there for you when times are hard. The first thing I tell founders I invest in is: always do what's best for your business, get advice, but go with your own gut.

Sandeep (@sandeepg33k) 's Twitter Profile Photo

when i was building the v0 of Bug0 AI agent, in the early days our AI agent sometimes wandered off and started testing competitor websites 😂 AI does weird stuff have guardrails from day 1

Sandeep (@sandeepg33k) 's Twitter Profile Photo

I have found that Grok Code Fast 1 (grok-code-fast-1) is a very capable model for multi step agentic loops. You can achieve complex tasks without breaking the bank. but `gemini-2.5-flash` is way faster, in some cases up to 50% faster (it costs $0.10 extra per million tokens).

Fazle Rahman (@fazlerocks) 's Twitter Profile Photo

We get ~1k pageviews a day on our new Bug0 site. Traffic isn’t the issue. Indexed on Google. Showing up in AI summaries. The issue: conversions. CTAs at 6% (good for B2B) but only 1–2 demos booked. Some days, none. Why? Nobody reads anymore. Landing pages all look the

Sandeep (@sandeepg33k) 's Twitter Profile Photo

since we started building Bug0 we’ve seen so many moments where a bug was just about to be shipped… but Bug0 caught it. why does this work? traditional browser testing is broken for one simple reason: fast moving teams ship so fast that the tests are outdated the very next

since we started building <a href="/bug0inc/">Bug0</a> we’ve seen so many moments where a bug was just about to be shipped… but Bug0 caught it.

why does this work?

traditional browser testing is broken for one simple reason: fast moving teams ship so fast that the tests are outdated the very next
Sandeep (@sandeepg33k) 's Twitter Profile Photo

In long agentic loops, adding a simple self-check like `doesActionAdvanceUsTowardsGoal` massively improves results. The AI takes fewer wrong turns, and when it does, it often flags false, giving you a chance to catch it and retry the step correctly.

In long agentic loops, adding a simple self-check like `doesActionAdvanceUsTowardsGoal` massively improves results.

The AI takes fewer wrong turns, and when it does, it often flags false, giving you a chance to catch it and retry the step correctly.
Sandeep (@sandeepg33k) 's Twitter Profile Photo

recently Bug0 AI caught a bug that regular playwright missed. instead of calling .clear() and .fill(), it acted like a real human - pressed cmd + A then typed. ..that subtle difference exposed the bug. browser tests need more than automation. They need to mimic humans!

Cloudflare (@cloudflare) 's Twitter Profile Photo

Today, we’re launching Cloudflare Email Service. Send and receive email directly from your Workers with native bindings—no API keys needed. cfl.re/4mx6fbO #BirthdayWeek

Fazle Rahman (@fazlerocks) 's Twitter Profile Photo

We're wasting our most talented engineers on a fundamentally broken process. We ask them to build innovative products, then saddle them with writing and maintaining brittle E2E tests. Every hour an engineer spends fixing a cy.get() selector or a flaky Playwright test is an hour

Fazle Rahman (@fazlerocks) 's Twitter Profile Photo

Super impressed with Codex so far. I’m using the Codex plugin in VS Code, and gpt-5-codex-low has been more than enough. This is the first time something’s actually outperformed Claude Sonnet 3.5/4 for me. If it keeps working this well, I might just cancel my Copilot and