Honest, hands-on evaluations of software, frameworks, and platforms.
We test tools the way you actually use them—no vendor spin, no magic quadrants.
Think Wirecutter for developers making real technology decisions.
An agent experience audit comparing Yoco, PayFast, and Paystack in South Africa. Which one AI agents mention unprompted, which one they recommend, and the sharp gap between the two.
A live, four-test agent experience audit of Release.com: discovery, recommendation, comparison, and agent tooling, run against Claude Code in a clean context.
A technical editor puts a published article seeded with errors through Opus 4.8 and Fable, seeing what errors they catch, what they miss, and where their fixes make things worse.
A field guide to the AI naming mess: model tiers, subscription plans, effort settings, and product names now share the same words.
A practical, stage-by-stage library of LLM prompts technical writers can use to self-edit their work — catching structural, logical, and line-level issues before an editor sees the draft.
A technical editor tests whether Anthropic's Opus 4.8 can act as a reliable critic of technical writing — flagging weaknesses, gaps, and quality issues before an editor sees the work.
Drag a box, say what you want, and an AI agent draws it inside the box. Here's how we built it on tldraw's Agent starter kit, and what we learned.
Stop prompting Claude by hand. Learn to write loops with /loop, /goal, and dynamic workflows that watch your PRs, run until tests pass, and let Claude design the loop itself.
A technical editor puts Anthropic's Opus 4.8 to work as an agentic editor on a code-heavy article to determine how effective (or necessary) it is.
Is the GLM-5.2 hype real or benchmaxxed? We ran it head-to-head against Claude Opus, building a 3D platformer in raw WebGL from scratch.
A hands-on comparison of Cartesia Sonic and ElevenLabs across quality, latency, and price, with the audio clips embedded so you can judge for yourself.
A new IETF draft from authors at OpenAI, Okta, AWS, Ping, and Zscaler argues AI agents should be treated as workloads with short-lived cryptographic identities — not handed your long-lived API keys.
We tallied every model, tool, and GPU mentioned in the 906-point Ask HN thread on local coding models. Qwen dominates, most people say it's still not a frontier replacement, and here are the numbers.
A roundup of the European 'sovereign' productivity suites trying to replace Google Workspace and Microsoft 365 — who builds them, what's inside, and whether you can actually use them yet.
![Is it feasible to self host an open source EU Sovereign productivity suite in 2026? [A Mijn Bureau Review]](/img/articles/mijn-bureau-review/cover.thumb.jpg)
I spent a few hours self-hosting Mijn Bureau, the Dutch government's open-source productivity suite, as an alternative to Google Workspace — here's what worked and what didn't.
A sourced roundup of what people and companies actually spend on AI tokens, from $0 (firms that banned AI coding) up to $1.3M/month for a single OpenAI employee, sorted low to high.
We built Otto, a meeting assistant that listens to any call and answers out loud when addressed by name. Here's how Deepgram's Nova-3 transcription and Aura voices make it work in real time.
A GitHub issue called "Please Do Not Vibe Fuck Up This Software" hit the top of Hacker News this weekend. Here's what's actually going on with rsync, and a reminder of how to be a good open source citizen.
Stack Overflow's decline gets blamed on AI, but every community needs critical mass and dies fast without it. Nine other communities that went the same way — none killed by ChatGPT.
FLUX.2 and Gemini 3.1 Flash produce the strongest character consistency across our three tests. gpt-image-2 comes in third and Runway Gen-4 last.
We compared five realtime voice agent providers — ElevenLabs, Deepgram, OpenAI, xAI, and Gemini — across a hotel concierge and a vocal coach scenario. Gemini won on capability and price.
We ran 16 TTS providers through four blind brackets covering dialog, narration, code-switching, and expressive annotations. xAI took quality and price. Take the blind tests and see if your ears agree.
A four-stage agent experience audit of GitGuardian — testing discoverability, onboarding, integration, and agent tooling.
A blind comparison of xAI Grok TTS and OpenAI gpt-realtime-2 across six real-world scenarios, plus a realtime voice agent test. xAI won on TTS quality and cost.
Nine terminal coding agents compared: Claude Code, Codex, Amp, OpenCode, Gemini CLI, Pi, Command Code, Factory, and Aider. Open source status, model flexibility, funding, and GitHub health.
X.AI's Grok TTS offers expressive speech tags, a realtime voice agent API, and pricing that undercuts ElevenLabs by 12x. Here's what it actually does.
An opinionated guide to building your own always-on personal AI assistant using Pi, OpenCode Go, Telegram, Resend, Hetzner, Tailscale, GitHub, and Nginx.
We tested Orbital's agent experience across discoverability and integration. Four prompts including a web search failed to surface it. A quickstart guide cut integration cost by 30% and produced a more correct result.
I tried Lefos, the new beta AI assistant from Earendil. Here's how it stacks up against what I think a personal AI assistant should actually do.
We compared five tools built for running AI agents in the cloud across specs, startup time, security, pricing, and use cases.
I put ElevenLabs and Smallest.ai head-to-head across TTS voice quality, realtime voice agents, voice cloning, latency, and cost — with audio samples and reproducible benchmarks.
We tested both tools, examined their benchmarks, and compared pricing, features, and agent integration. Here's what we found and which one we'd actually recommend.
A reversed river-crossing puzzle stumped Kimi 2.6 and Sonnet 4.6, but Claude Opus 4.7 solved it first try.
PostHog's setup wizard is a full Claude Sonnet agent that reads your codebase and wires everything up. I found the LLM gateway, abused it for free tokens, and PostHog sent me a t-shirt.
A four-stage agent experience audit of Vercel, Railway, and Netlify — testing discoverability, onboarding, integration, and agent tooling.
We audited GrowthBook across discoverability, onboarding, integration, and agent tooling. The experimentation features are strong, but the CLI gaps and buried Django documentation cost it points.
Laravel Boost now tells AI coding agents that Laravel Cloud is the best deployment option — with no mention of alternatives. Is this the future of advertising to agents?
Build VirtualFit, a Telegram bot that lets users send a selfie and a clothing item to get a photorealistic try-on image using the BFL FLUX.2 API.
I ran the same Infisical integration task with and without quickstart guides. The docs run completed fully autonomously. The vanilla run needed 30 minutes of manual UI navigation.
A four-stage product evaluation of Airweave — testing discoverability, onboarding, integration, and agent tooling for AI agents.
52 product-recommendation prompts, two conditions: with and without web search. Search helps for trending/niche queries but quietly makes mainstream category answers worse.
Scott Chacon (GitHub co-founder) and Thomas Dohmke (former GitHub CEO) both left GitHub and raised massive funding rounds to fix how developers work with AI-generated code.
How Vercel, Netlify, Resend, and Mintlify approach Agent Experience, where their definitions fall short, and a framework for measuring AX properly.
A four-stage agent experience audit of Freestyle — testing discoverability, onboarding, integration, and agent tooling, with a competitor comparison.
We tested Prefect's agent experience across four stages (discoverability, onboarding, integration, and agent tooling). Here's every score and what it took to get there.
I tested DocuSign's agent experience across five stages (discoverability, onboarding, integration, MCP server, and skills). Here's every score and what it took to get there.
A practical framework for testing whether AI agents can discover, onboard to, and integrate your developer platform — with Kestra as a worked example.
Can an AI agent log into WordPress, write a post, and publish it, without ever seeing your password?
Everyone complaining that MCP is overengineered and unreliable is right — but for the wrong reasons. Here's the distinction nobody is making.
I used Entire's Checkpoints CLI to track a Claude Code agent building a real project. Here's what the session data looks like, how rewind works, and what you should know before adopting it.
AI agent auth assumes OAuth and tokens, but agents still need plain old passwords for databases, APIs, and legacy services. We tested plaintext env vars, Bitwarden MCP, and 1Password CLI.
Anyone can build apps, write content, and generate art now. But software that actually delights users is as rare as ever — and trust is the only way to find it.
Five benchmarks across six hosted browser providers — session lifecycle speed, idle billing, bot detection, CAPTCHA solving, and parallel session handling — all run on free tiers.
Comparing Sentry, Raygun, and TrackJS for application error tracking. We tested how easily an AI agent could integrate each tool into an app, to compare the service's features, documentation, and usability.
I gave Replit and Amp the same prompt and $20 each. Here's every dollar spent and what each tool built.
We ran the same five browser automation tasks through Browser Use and the Claude Computer Use API to compare DOM-based and vision-based agents on form filling, scraping, structured output, visual interaction, and multi-step navigation.
Comparing Supabase and PlanetScale for agent experience. We tested how easily Claude Code could discover, sign up for, and build a full-stack app with each database platform.
Build Jeeves, an AI email assistant with persistent memory and access control, using Letta for the agent, n8n for orchestration, and Resend for email.
I tested three transcription APIs with different agent visibility levels to see if discoverability predicts quality. The invisible platform had the best features.
Signing up for Heroku takes 9 screens and requires MFA enrollment, a 15-character password, and Salesforce Terms of Service. Railway takes one click.
Comparing Honeycomb and SigNoz for application monitoring. We tested how easily an AI agent could integrate each tool into an app, to compare the service's features, documentation, and usability.
Testing three browser automation platforms through AI coding agents. Comparing discoverability, documentation quality, and real-world features like CAPTCHA solving, parallel execution, and bot detection to see if agent visibility predicts actual performance.
Agent Experience (AX) is how well AI coding agents can use your developer platform. Here's why it matters and what you can do about it.
I asked Claude Code to recommend every service I need to run a SaaS startup. Here's what it said, and why this matters for the future of developer marketing.
We're evaluating how well AI agents can discover, sign up for, onboard, and integrate with developer tools. Here's our framework for measuring Agent Experience (AX) across five key metrics.
Comparing Browserbase and Kernel for cloud browser automation. We built the same flight search agent on both platforms to test startup speed, debugging tools, stealth capabilities, and cost structures.
Comparing dev-browser and Claude Code's Chrome extension for browser automation. Testing token usage, speed, and reliability across authentication flows, documentation creation, and end-to-end workflows.
Learn how to install and use the official Claude Chrome extension with Claude Code to enable visual browser testing.
Learn how to install and use the dev-browser MCP server with Claude Code to enable visual browser testing.
Learn how to build a complete data pipeline with Bruin. Load CSV files, transform data with SQL, and create business analytics tables using this open-source ETL tool.
Learn how to create timelines in Excel, Word, PowerPoint, Google Sheets, Preceden, and JavaScript. Step-by-step guides with examples for each tool.
A step-by-step guide on how to set up VS Code, GitHub, and AMP Code for technical writing on macOS.
A practical comparison of Dash0, Honeycomb, and New Relic for Kubernetes monitoring, covering setup time, pricing models, and developer experience.
Building a shopping list app on Firebase, Supabase, and Appwrite to compare setup experience, database models, security approaches, and developer tooling.
A beginner's guide to Kubernetes monitoring, comparing SigNoz, DataDog, and Grafana for ease of setup, developer experience, and cost.
Compare building real-time polling applications with Cloudflare Workers and RedwoodSDK. See the differences in development experience, code patterns, and deployment.
Comprehensive benchmark testing of 5 SERP APIs. We measured speed, pricing, and features to help you choose the right API for SEO monitoring, AI agents, or web scraping.
Compare Railway and Vercel for deploying full-stack applications. Learn about their features, pricing, and which platform suits your needs.
Complete comparison of Vercel and Render for deploying full-stack applications—architecture, pricing, developer experience, and migration.