Tools I Use

Claude Code has become my primary tool for serious coding work, particularly for complex refactors where the Opus 4.7 model's 1 million token context window genuinely changes what you can ask for in a single pass. The remote control feature has reshaped how I work, letting me direct sessions from my phone while walking outside and still producing code that holds up under review. It runs slower than inline completions, but the quality of the output has consistently been worth the wait.

Cursor is the editor I reach for when I need speed: tab completions that actually match my style, surgical edits on single files, and quick side-project prototypes. The Composer 2 mode trained on Kimi 2.5 provides generous usage limits, which makes it ideal for the kind of fast zero-to-one iteration where you want to test ideas cheaply before committing to a direction. My simple division of labour is Cursor for typing speed and Claude Code for thinking speed.

Google AI Studio's Build feature is my go-to for quickly prototyping apps, websites, and UI ideas. When I have a rough concept for a landing page, a small web tool, or a layout I want to try, I describe it in a prompt and see a working version rendered back inside a minute. The free tier gives plenty of room to iterate, and the speed advantage keeps bringing me back whenever a new idea needs a fast visual sanity check.

Claude.ai is my everyday assistant for general AI work, spanning both personal projects and daily business tasks. The combination of strong reasoning and genuinely long context makes it the first tab I open when I need to think through a problem rather than just autocomplete a sentence. When I need fresh web sources I switch to Perplexity, and when the end product is code itself I move over to Claude Code.

Perplexity is the tool I use for serious research work, because its citations reliably surface sources that Claude and ChatGPT would never show me. It is also useful for quickly testing ideas against the open web, though the Pro tier feels expensive unless deep research is part of your weekly routine. The free tier is sufficient for most people to understand the pattern and decide whether they need more.

Google's AI Search mode has become my preferred tool for quick research queries, particularly when I am away from my desk and only have my phone. The integration with Search delivers citation-backed answers faster than any other assistant I have tested, which matters more than I expected. For deeper research I still reach for Perplexity, but for fast on-the-go answers, Gemini is the clear winner on both speed and quality.

Grok has become my default for anything that requires current X data, and the live search integration is genuinely useful for trend spotting, tracking what is going viral, and understanding what founders are complaining about in real time. It is far more reliable than scraping X directly, and the reasoning quality is strong for the price. For longer-form work I still prefer Claude, but Grok wins on anything time-sensitive.

Hermes Agent is my preferred framework over OpenClaw for the kind of self-hosted automation I rely on day to day. We run it on a single affordable Hetzner VPS, where it handles scheduled operational tasks, cross-service integrations, and internal monitoring jobs without complaint. The skill framework is the feature that made it click for me: each scheduled task is written as a declarative skill definition rather than a brittle shell script, which makes long-term maintenance dramatically easier.

Firecrawl has become my primary scraping tool, and I recently upgraded to the 100,000-credit monthly plan, which gives us enough headroom that I no longer have to watch the usage meter during research-heavy weeks. The batch API is the feature that matters most when I need to pull structured data from a long list of URLs in one pass. It is not the cheapest option at scale, but the combination of JavaScript rendering and clean markdown output has been worth every pound.

Convex powers the backend for myaiguide.co, and the combination of scheduled functions, realtime queries, and strictly typed mutations lets me ship features faster than any other backend stack I have used. The free tier is genuinely generous, so costs stay predictable as you grow. The only meaningful constraint worth planning around is the 10-minute cap on actions, which you will want to route longer jobs off of.

Vercel is where every frontend I build eventually lives, thanks to its per-PR preview deploys, clean redirect syntax in next.config, and ISR with revalidateTag support. The git-push auto-deploy has been unreliable this year, so I now always follow every push with an explicit `vercel deploy --prod --yes`. Everything else about the platform works as advertised, which is unusually rare in modern developer tooling.

Tavily was my go-to research API before Firecrawl upgraded its offering. The search and extract endpoints are clean and dependable, and the API surface is easy to reason about. I am currently migrating our pipelines to Firecrawl now that we have the 100,000-credit plan, but Tavily's free tier remains a valuable backup for any future scenario where we need to split research calls across multiple providers.

DataForSEO is my primary source for keyword and SERP data, and the combination of coverage, reliability, and reasonable pricing has kept me using it over every mainstream alternative I have tried. The raw API gives me exactly the fields I need without the bloated UI layer that most SEO platforms insist on, and the Task POST mode keeps costs sensible when I am running larger batches. For serious research work, it is the only source I fully trust.

ClickUp is the operational hub for every project I run, covering tasks, review queues, and daily planning in a single workspace. The REST API is reliable enough that I wire notifications to it directly from almost every internal system without needing any middleware in between.

PostHog provides our product analytics layer, and the HogQL query language has been genuinely useful because I can slice user behaviour directly without first building a dashboard. The EU-hosted instance satisfies our GDPR requirements, and the session replay feature has already caught a bug that no Lighthouse audit would have flagged. Pricing stays affordable until you hit serious scale on session recordings.

Beehiiv is the platform I use to host the myaiguide.co newsletter, and the dashboard has quietly become one of my favourite pieces of software to open. The API accepts drafts programmatically, which is the feature I lean on most, and the editorial workflow around scheduling and A/B testing subject lines is genuinely well thought through. For anyone serious about owning their newsletter and not renting it from Substack, it is the choice I keep coming back to.