The 5 AI Tools I Actually Use to Run My Business (2026)

[Operator’s read] I run two businesses: a personal AI-consulting brand (alejandrorioja.com) and Pickleland, a pickleball facility in Pflugerville, TX. Different contexts, different audiences, different ops. These five tools run both. I’m not listing them because they’re trendy; I’m listing them because I’ve deleted their replacements.

Table of contents

Open Table of contents

• 1. Claude — the operator layer
• 2. Cursor — where the TypeScript gets written
• 3. Airtable — the data backbone
• 4. Kit — newsletter and email automation
• 5. Cloudflare Workers — where the agents live
• What I cut and why
• The operator’s bottom line

1. Claude — the operator layer

Claude (via Claude Code and the Anthropic SDK) is the brain of everything that moves. I use it in three modes:

Claude Code is my daily driver for development. I write TypeScript, build agents, debug infrastructure issues, and manage content — all from the Claude Code interface. It’s not just autocomplete; it’s a collaborator that can read a 500-line file, understand intent, and propose a refactor I hadn’t considered.

The Anthropic SDK powers every agent I’ve built. My newsletter agent, my Facebook ads skill, my content pipeline, my OG card generator — all Claude on the backend. The model quality is high enough that I trust first drafts about 85% of the time.

Claude’s voice and brand judgment is underrated. When I’m writing something that needs to sound like me, I’ve found Claude + a detailed system prompt outperforms every other model I’ve tested. The trick is a specific, opinionated system prompt — not “write in a casual tone” but “write like Alejandro: direct, practitioner, no hype, numbered, first-person, with honest caveats.”

I pay for Claude Max. It’s the most-used subscription I have, and the ROI is not close.

2. Cursor — where the TypeScript gets written

Cursor is the IDE. I switched from VS Code about a year ago and haven’t looked back.

The tab completion is fast enough that it genuinely changes how I write code — I think at a higher altitude and let Cursor handle the syntactic boilerplate. The diff view for AI suggestions is clean. The multi-file context window means I can ask it to update a function and it updates the callers too.

I don’t use Cursor for architecture decisions. I still sketch those on paper or in Claude. But once the design is clear, Cursor is the fastest path from design to running TypeScript.

The biggest unlock: Cursor + Claude Code in parallel. I use Claude Code for high-level planning and agent orchestration; I use Cursor for the implementation detail work. They don’t conflict — they cover different altitudes.

3. Airtable — the data backbone

Every AI agent I run needs a place to read from and write to. That place is Airtable.

Here’s what I use it for across both businesses:

• Content queue — posts and newsletter topics in progress, with status tracking
• Booking records — Pickleland court reservations synced from the booking system
• Affiliate link catalog — 105+ slugs with metadata the content agent reads at generation time
• Agent audit log — what ran, when, what it produced, any errors

The API is clean and fast. Airtable is not a database for high-throughput workloads — but for agent side-tables, review queues, and human-in-the-loop approval workflows, it’s exactly the right tool. The visual interface means I can inspect any table without writing a query.

The alternative I tried: Notion databases. The Notion API is slower and the data model is clunkier for agent reads. Airtable wins for agent-adjacent data.

4. Kit — newsletter and email automation

I switched to Kit (formerly ConvertKit) for one reason: the API is actually good.

Most email platforms treat their API as an afterthought. Kit treats it as a first-class product. I can create broadcasts, schedule sends, segment by tag, and read analytics — all programmatically. My newsletter agent does all of this without me touching the composer.

Kit-specific things I use:

• Broadcasts API — my agent creates scheduled broadcasts programmatically every week
• Subscriber tagging — I tag subscribers by behavior (opened last 5 sends = “engaged”; hasn’t opened in 60 days = “at-risk”) and my agent targets segments accordingly
• Forms + landing pages — clean, fast-loading, no-code. I don’t mess with these programmatically; they just work.

If you’re on Mailchimp or a legacy platform: the migration is worth it. Mailchimp’s API requires three extra calls to do what Kit does in one.

5. Cloudflare Workers — where the agents live

Every scheduled agent runs on Cloudflare Workers. The pitch: global edge deployment, zero cold starts on the free tier, and a cron trigger system that actually works.

My agents don’t need a server. They need a scheduled function that runs reliably, can make external API calls, and costs close to nothing at my scale. Workers is the answer.

What I have running on Workers:

• Content pipeline — generates EN post, fans out to 12 translations, generates OG card
• Newsletter agent — drafts and schedules the weekly send
• Facebook ads monitor — reads performance, flags underperformers, notifies me
• Pickleland occupancy reporter — reads booking data, sends me a daily summary

Total monthly cost for all of this: ~$5. That’s the paid Workers plan. The agents run reliably on the cron schedule; I’ve had one failure in six months (a DNS issue on Meta’s side, not mine).

What I cut and why

Zapier — replaced by Workers + the respective platform APIs directly. Zapier adds latency, costs more at scale, and has a ceiling that Workers doesn’t.

ChatGPT — Claude’s context window, tool use, and system prompt quality are better for the operator use case. I keep a ChatGPT tab for quick web searches but don’t build on it.

Webflow — moved my site to Astro + Cloudflare Pages. More control, better performance, build process I can script against.

Grammarly — Claude does everything Grammarly does and keeps my voice better.

The operator’s bottom line

The five tools above are not the newest or the most-discussed. They’re the ones that held up under daily production use across two different businesses. Before adding a new tool to your stack, ask: which of these five could do this job? You’ll be surprised how often the answer is “one of them already can.”