A publication for solo founders, indie hackers, and small-team builders. Honest tool reviews. Side-by-side comparisons. Every numeric claim linked to a primary source. No affiliate ranking. No sponsored slots in reviews.
Most AI content is written for buying committees at companies with budgets you don't have. Most AI tool reviews are paid placements dressed as opinions. Most newsletters summarize TechCrunch.
SoloBuilder is the publication I wished existed when I was deciding whether to pay for Cursor, Windsurf, or both. It's written for one specific reader: the solo founder or small-team builder figuring out which AI tools actually move the needle on their constraints — not the constraints of a 500-person engineering org.
If that's you, the rest of this page is the case for why this is worth your inbox space.
When I write "Cursor's Pro plan is $20/month," the link goes to cursor.com/pricing. Not a screenshot from 2024. Not a paraphrase from another newsletter. The actual page, fetched the day the article shipped. If a price changes, the article gets a [revised: date] tag.
When I review a tool, the question I'm answering is: "Is this worth the money for one person shipping a B2B SaaS?" Not "does it support SAML SSO." Not "does it scale to 10,000 seats." Solo-founder constraints — pricing, time-to-first-value, cancellation friction — are the rubric.
APEX is an autonomous research engine I built. For every comparison article, it queries 35+ sources, scores them for authority and recency, drafts a synthesis, runs an 8-vector adversarial critique against its own draft, argues both sides of every verdict, and grades the result against a 12-criterion rubric. Then I edit it, cut the filler, and put my name on it. The byline is a person; the rigor is a system.
The structure of every issue is the same so you know exactly what you're getting:
No filler. No threads disguised as bullets. The whole thing reads in five minutes.
The system behind every cited claim. Built by me. Runs on my server. Outputs are public — every article on this site went through it.
Every article starts with a falsifiable claim. "Is Cursor worth $20/month for a solo dev shipping a B2B SaaS?" Not "Cursor: a comprehensive review." The question constrains everything that follows.
APEX queries the open web through Firecrawl, prioritizing pricing pages, official docs, GitHub issues, changelogs, and benchmark repos. Each source gets scored on authority (gov/official > established media > blogs > forums) and recency (the last 90 days weighs more than 2-year-old reviews).
APEX writes a synthesis. Then it runs an 8-vector adversarial critique against its own draft — checking for overclaiming, weak evidence, missing counterarguments, verdict-evidence mismatch. It argues the bull case and the bear case for the verdict. It grades the whole thing on a 12-criterion rubric. Sections scoring below 7.0 get auto-revised before I see the draft.
I read the APEX draft, cut filler, add nuance, fix tone, and put my name on the byline. Citations stay visible in the final article. Inferences are tagged [INFERENCE — confidence: H/M/L]. Estimates are tagged [ESTIMATE — verify: URL]. Nothing speculative ships unmarked.
The full pipeline takes 5–15 minutes per article. Same job done by hand: 4–16 hours. The reason I built APEX wasn't to write faster. It was to write more honestly than I could on my own — because the system's first job is to attack its own draft and tell me where it's weak.
The tools behind solobuilder.ai, APEX, and everything else. Updated quarterly. No affiliate links in this list — these are just what I run.
Primary development environment. Where solobuilder.ai gets built and APEX gets debugged.
Strategic conversations, architecture decisions, copy review.
solobuilder.ai is built on this. Vercel for hosting.
Styling system. Editorial palette, three Google Fonts.
Every blog article is an MDX file. Git is the CMS.
Newsletter delivery and subscriber management.
APEX's web scraping layer. Handles the 35-source queries.
Second brain. Build logs, strategic docs, the APEX Knowledge Vault.
DNS, custom email setup, SPF/DKIM/DMARC.
Personal AI agent on WhatsApp/Telegram for internal automation.
The ones that come up in reader replies, answered plainly.
Start with one general AI assistant — Claude.ai or ChatGPT, $20/month — and use it daily for two to four weeks before adding anything else. Use it for writing, drafting emails, structuring decisions, pressure-testing ideas. Build the habit of reaching for it before any task that involves a blank page.
Don't add a second tool until you have a specific, repeated bottleneck the assistant can't solve. Wasting time on meeting notes? Add a transcription tool. Drowning in form-to-app data entry? Add Zapier or Make. Identify the pain, then add the tool. Not the other way around.
The founders who skip the assistant phase and jump straight to agent builders, automation suites, or specialist tools consistently hit friction that makes them conclude AI isn't ready. It's not that AI isn't ready. The starting point was wrong.
$100–500 per month covers most solo-founder AI stacks in 2026. Typical breakdown for a founder six months in:
• $20–40/month for one general assistant (Claude Pro or ChatGPT Plus) • $20–30/month for one coding tool if you're building software (Cursor, Claude Code, or Copilot) • $20–50/month for one automation platform (Zapier, Make, or n8n self-hosted) • $0–100/month in API credits for any custom workflows you build • $50–200/month in specialist tools that solve specific bottlenecks (transcription, email, analytics)
If you're spending more than $500/month, you probably have overlap — multiple tools doing the same job — and you should audit. If you're spending less than $50/month, you're either at the very beginning or being too cautious. The right answer is rarely the cheapest stack. It's the stack that buys back time you can spend on revenue work.
Three tests, in order:
1. Identify the specific manual task the tool would replace. "It automates the follow-up emails I write after every discovery call, and setup takes two hours" is clear. "It enables AI-powered workflow orchestration" is marketing. If you can't describe the replaced task in plain language, you don't yet know whether the tool fits.
2. Trial it for 14 days against the actual problem. Not against demo data. Not against an example workflow from the vendor's docs. Your real data, your real workflow. If the tool doesn't deliver value within the trial, it won't deliver value after you start paying.
3. Calculate hours saved per week. A tool that saves 2+ hours per week pays for itself at $50/month. A tool that saves 30 minutes per week probably doesn't, even at $10/month — the cognitive overhead of remembering it exists costs more than the time it saves.
Anything I review on this site has passed all three tests in production for at least 30 days before the article ships.
Yes. Three things specifically:
Models got materially better. The gap between mid-2025 and early 2026 is larger than the gap between late 2024 and mid-2025. Tasks that produced obvious hallucinations a year ago — code refactoring across files, multi-step research, structured output — now work reliably. If you bounced off in 2024 or early 2025, it's worth one more honest evaluation.
Tooling caught up. The first wave of AI tools were thin wrappers over OpenAI's API. The current wave have real product layers — context management, persistent memory, tool integrations, structured workflows. That's the difference between a chat box and an actual product.
Pricing rationalized. Most useful AI tools now sit in the $20–50/month range with real free tiers. The era of "unlimited GPT-4 for $5 or $200/month enterprise" is over.
If the failure mode last time was "the tool was too dumb," try again. If the failure mode was "I never figured out what to use it for," the tools haven't fixed that — that's still on you. This site's main job is helping with the second one.
Pull every recurring AI subscription into a spreadsheet. For each one, write down:
• What specific task does this replace? Not "helps with content" — write the actual task. "Drafts the LinkedIn post that goes out Monday morning." • When did I last use it for that task? If it's been more than two weeks, that's a signal. • What would happen if I cancelled it tomorrow? If the answer is "I'd switch to Claude/ChatGPT and barely notice," cancel it. • Is it overlapping with another tool I'm paying for? Two tools that both write social posts. Two tools that both transcribe meetings. Pick one.
Most solo founders I talk to are running 2–3 tools they haven't logged into in a month. Cancelling those isn't austerity — it's the only honest way to know what's actually delivering value. After the audit, your remaining stack becomes evidence: these are the tools I genuinely use, and the others were aspirational.
The rule: a tool earns its monthly recurring slot or it gets cancelled. No tool gets sentimentality.
Not all three. Pick one and run it in production for at least 30 days before evaluating.
The short version:
• Cursor — most polished IDE experience. Best if you want a familiar VS Code interface with AI woven through every action. $20/month. • Windsurf — newer, more agentic feel, strong for multi-file edits and large refactors. $15/month. • Claude Code — terminal-based, best for agentic workflows where you want Claude to actually run commands, edit files, and verify changes itself. Bundled with Claude.ai Pro.
For a solo developer shipping a B2B SaaS: Claude Code if you're comfortable in a terminal and want the most autonomous experience. Cursor if you want a traditional IDE with AI assistance.
The deeper version — specific tasks each one wins on, the tradeoffs, the cost-per-feature analysis — is on the blog.
Pick three sources and ignore everything else. The discipline is in what you cut, not what you read.
A reasonable trio:
• One tool-focused newsletter for honest reviews and comparisons. (This site, ideally.) • One operator-focused source — someone like Lenny Rachitsky, Pieter Levels, or another founder publishing what's actually working in their business. Not a news aggregator. • One technical changelog feed for the specific tools you use — Cursor releases, Anthropic docs, Vercel changelog. Not generic AI news.
Mute everything else. Mute "AI Twitter." Mute the tool launch announcements that aren't tools you're actively considering. Mute the hot takes on which model is best this week.
The best signal in this space is dramatically lower-volume than the noise. If you're spending more than 30 minutes a day on AI news, you're consuming, not building.
AI tools wait for you to ask. AI agents are given a goal and figure out the steps themselves.
A tool: you write a prompt, it returns a draft. Every action requires you in the loop.
An agent: you say "draft this week's newsletter from the three articles I shipped and post it to Beehiiv when it's ready for review." The agent figures out the sequence, calls the tools needed, handles errors, reports back.
For most solo founders in 2026: tools are your daily driver, agents are an experiment. Tools are mature, well-documented, reliable. Agents are improving fast but still require careful constraint — they need explicit permissions for what they can do autonomously vs what requires approval, and the failure modes are harder to debug.
Start with tools. Add an agent only when you have a multi-step workflow you've already manually performed enough times to write down each step. The worst time to introduce an agent is when you don't yet know what the workflow looks like.