# The AI Quick Start for Launching a Business: What to Install, What to Pay For, and What You Will Not Own

One subscription and a terminal will do the market research, write the site, generate the brand assets, and run the audits. Here is the stack, the current prices, and the copyright limit nobody mentions.

Author: J.A. Watte
Published: August 6, 2026
Source: https://jwatte.com/blog/ai-quick-start-for-launching-a-business/

---

The work of launching a business used to be gated by access to specialists. Market research meant a consultant. A website meant a developer. A logo meant a designer. Structured data meant an agency retainer.

Most of that gate is gone, and the replacement costs less than a phone plan. What has not changed is that you still need to know what to ask for, and there are two or three places where the AI version produces something you do not legally own in the way you assume.

This is the practical stack, the current prices, and the honest limits.

## The one decision that matters

Almost everyone starts by picking a chat product. The more consequential decision is whether you work in a **browser chat window** or in a **terminal agent that can read and write files on your machine**.

For launching a business the terminal agent wins, and it is not close. A chat window gives you text you then paste somewhere. A terminal agent builds the site, writes the files, runs the audit script, fixes what it finds, and deploys, without you being the copy-paste layer between steps.

The cost is one evening of setup discomfort if you have never used a terminal. That is the entire cost, and it is worth paying.

I wrote the fresh-machine setup as [AI Terminal Kickstart](/blog/blog-ai-terminal-kickstart/), with runnable install scripts for each platform:

- Linux: [/downloads/start-ai-terminal-kickstart.sh](/downloads/start-ai-terminal-kickstart.sh)
- macOS: [/downloads/start-ai-terminal-kickstart-macos.sh](/downloads/start-ai-terminal-kickstart-macos.sh)
- Windows: [/downloads/Start-AITerminalKickstart.ps1](/downloads/Start-AITerminalKickstart.ps1)

## What it costs

Claude's current plan tiers, read from the pricing page today:

| Plan | Price | Claude Code included |
|---|---|---|
| Free | $0 | No |
| Pro | $17 / month billed annually, $20 monthly | Yes |
| Max | From $100 / month | Yes |
| Team standard seat | $20 / seat / month billed annually, $25 monthly | Yes |
| Team premium seat | $100 / seat / month billed annually, $125 monthly | Yes |

The relevant fact for a founder: **the terminal agent is included at the $17 tier.** You do not need the $100 plan to do the work in this post. Move up only when you hit usage limits often enough to be annoyed, which for launch-shaped work is unlikely in the first month.

If you would rather pay per token than per month, the API is metered separately. Current published rates are $5 per million input tokens and $25 per million output for Claude Opus 5, $3 and $15 for Claude Sonnet 5, and $1 and $5 for Claude Haiku 4.5. For the intermittent bursts of a launch, a subscription is almost always cheaper than metered API usage; the API makes sense when you are building something that calls a model in production.

**On the alternatives.** I could not load OpenAI's consumer pricing page in this pass (it returned a 403 to my fetch), so rather than repeat a number from memory I will simply say: the competitors are in the same range, competitive terminal agents exist, and my comparison of them is in [which terminal AI coding agent](/blog/which-terminal-ai-coding-agent-claude-codex-gemini/). Check current pricing at the vendor rather than trusting any article for it.

## The five jobs, and how to actually ask

The difference between a useful launch stack and an expensive autocomplete is entirely in the asking. Five jobs, in order.

### 1. Market and name research

Ask it to do research, not to have opinions. The prompt shape that works: give it the specific market, the geography, and the constraint, then ask for what it can verify with links.

The mistake is asking "is this a good business idea." You will get agreeable prose. Ask instead for the ten competitors within your service area, what each one charges, what their sites do not say, and which questions customers ask that nobody answers well. That last one is your content plan and frequently your positioning.

Before you fall in love with a name, clear it. Not after the business cards. [Trademark Pre-Screen](/tools/trademark-prescreen/) runs the twenty-minute version, and the full cost and process picture is in [how to trademark a logo](/blog/how-to-trademark-a-logo-uspto-cost/). The federal filing fee is currently $350 per class, so a name collision discovered late is expensive in a way a name collision discovered early is not.

### 2. The site

Do not ask for "a website." Ask for a specific site with specific constraints, and give the model the standards you want met up front rather than auditing them back in later.

That is what [Single Site Gen](/tools/single-site-gen/) is for: it generates the full build prompt with the security headers, accessibility requirements, structured data, and bot posture already specified, so the first draft is close to correct instead of needing three rounds of remediation.

The practical rule: **specify the whole task in the first message.** Agents do their best work when the goal, the constraints, and the definition of done arrive together. Feeding requirements in one at a time across ten messages produces worse output and costs more.

### 3. Brand assets

Image models will produce a logo, a favicon set, hero images, and social cards in an afternoon. [Image Prompt Gen](/tools/image-prompt-gen/) writes the prompts and emits the post-processing commands; [the favicon and brand image walkthrough](/blog/blog-ai-image-generation-brand-assets/) covers the workflow end to end, and [ImageMagick for favicon sets](/blog/blog-imagemagick-favicon-brand-creation/) covers the conversion side.

**Now the part nobody tells you, and it is important.**

The US Copyright Office requires human authorship for copyright protection. Its Part 2 report on copyrightability, published 29 January 2025, addressed outputs created using generative AI, and its registration decisions have been consistent: purely AI-generated material is not registrable. The *Zarya of the Dawn* decision of 21 February 2023 refused registration for material created entirely through AI, and the *Théâtre D'opéra Spatial* review board decision of 5 September 2023 reached the same conclusion. Works combining human and AI-generated elements can receive protection **limited to the human contribution.**

So: a logo you generated from a text prompt and used unmodified is unlikely to be something you can register a copyright in.

**Copyright and trademark are different regimes**, and this is where I want to be careful rather than reassuring. Trademark protects a mark used in commerce to identify the source of goods or services; the test is distinctiveness and use, not authorship. Those are genuinely different questions. But I could not locate a primary USPTO source in this pass stating how AI generation affects the registrability of a mark, so **I am not going to tell you it is fine.** Treat it as an open question, and if the mark matters to your business, ask a trademark attorney before you build a brand on it rather than after.

The low-risk path, regardless: use the generated image as a starting point and do real human design work on top of it. That improves the logo anyway, and it puts you on the right side of the human-authorship line for whatever protection is available.

### 4. The structured data and the technical layer

This is where AI assistance is least glamorous and most valuable, because the work is precise, tedious, and easy to verify.

Ask it to write your `Organization` and `LocalBusiness` JSON-LD, your `robots.txt`, your `llms.txt`, and your DNS records for email authentication. Then **verify every one of them with a tool rather than trusting the output**, because this is exactly the category where a confident wrong answer looks identical to a right one.

- [Schema Validator](/tools/schema-validator/) for the JSON-LD
- [DNS Records Generator](/tools/dns-records-generator/) and [DNS / Email Auth Audit](/tools/dns-email-audit/) for SPF, DKIM, and DMARC
- [LLMs.txt Validator](/tools/llms-txt-validator/) for the spec compliance
- [Legal Pages Generator](/tools/legal-pages-generator/) for privacy, terms, and accessibility statements

### 5. The audit loop

Point the agent at the results of a real audit and let it fix things. Most tools on this site emit a paste-ready remediation prompt for exactly this: run the audit, copy the prompt, hand it to the agent, let it patch the code, re-run the audit.

The whole eight-dimension pass is in [the pre-launch audit loop](/blog/pre-launch-web-audit-loop/). Run it, feed the failures to the agent, re-run. That loop is the single highest-yield use of an AI subscription during a launch.

## Where AI will confidently mislead you

Five categories where I have watched it produce wrong answers with total assurance. Verify these against a primary source every time.

1. **Fees and filing costs.** Trademark fees, state filing fees, and program costs change and the training data is old. Check the agency page.
2. **Program eligibility.** Set-aside rules, certification requirements, and loan criteria are exactly the kind of thing that changed last year and reads plausibly either way.
3. **Whether a product still exists.** Platform products get retired and merged constantly, and a model will happily describe a workflow for something that shut down two years ago.
4. **Current pricing for anything.** Including the numbers in this post, which is why they carry a date and a link.
5. **Legal thresholds and deadlines.** The FinCEN beneficial ownership requirement, for example, reversed for domestic US companies in March 2025. A model trained before that will state the old rule with complete confidence.

The reflex worth building: when the answer is a number, a deadline, or an eligibility rule, ask for the source URL and open it. If it cannot produce one, treat the answer as a hypothesis.

## The whole stack, honestly totalled

- **AI subscription with a terminal agent:** $17 to $20 a month
- **Domain:** roughly $10 to $20 a year
- **Hosting:** $0, and see [the hosting cost comparison](/blog/serverless-hosting-cost-comparison-2026/)
- **Audit tools:** $0
- **Business formation:** state filing fee, which varies, plus $0 for the EIN
- **Optional, only if the name matters enough:** $350 per class for a federal trademark application

That is a real business launched for something close to the cost of a single hour of agency time. The gap between that and the $5,000 quote is not quality. It is that you did the asking yourself.

If you want the sequenced version of the whole thing rather than the tooling slice, that is [the observe, evaluate, launch playbook](/blog/observe-evaluate-launch-brand-playbook/). And the argument that this is a $20-a-month problem rather than a $2,000-a-month one is the entire subject of my book *The $20 Dollar Agency*.

## Fact-check notes and sources

- Claude plan tiers and prices, and Claude Code inclusion from the Pro tier upward, read 6 August 2026: [claude.com/pricing](https://claude.com/pricing)
- Claude API token pricing for Claude Opus 5, Claude Sonnet 5, and Claude Haiku 4.5: [Claude platform pricing](https://platform.claude.com/docs/en/pricing)
- US Copyright Office human authorship requirement, the Part 2 report on copyrightability published 29 January 2025, the *Zarya of the Dawn* decision of 21 February 2023, and the *Théâtre D'opéra Spatial* review board decision of 5 September 2023: [copyright.gov/ai](https://www.copyright.gov/ai/)
- Current federal trademark application fee of $350 per class, effective 18 January 2025: [USPTO trademark fee information](https://www.uspto.gov/trademarks/trademark-fee-information)
- FinCEN's March 2025 interim final rule exempting entities created in the United States from beneficial ownership reporting, cited above as an example of a reversed rule: [fincen.gov/boi](https://www.fincen.gov/boi)

*I was unable to load OpenAI's consumer pricing page during this pass and have therefore made no claim about its current prices. Ideogram's commercial-use terms were likewise not verified in this pass; read the terms for whichever image tool you use before shipping a logo, because commercial rights differ by plan on several of them. Prices and legal positions change; every figure above carries the date it was read. This post is informational, not legal advice, and the copyright and trademark discussion in particular is a summary of public agency positions rather than advice about your situation. No affiliation with any vendor named is implied.*

## Related reading

- [AI Terminal Kickstart: set up Claude Code, Codex, and friends on a fresh box](/blog/blog-ai-terminal-kickstart/)
- [The $50-a-month AI stack for a small business](/blog/blog-50-month-ai-stack-smb/)
- [How to generate favicons and brand images with Midjourney and Ideogram](/blog/blog-ai-image-generation-brand-assets/)
- [How to trademark a logo, and what it actually costs](/blog/how-to-trademark-a-logo-uspto-cost/)
- [The pre-launch audit loop](/blog/pre-launch-web-audit-loop/)


---

Canonical HTML: https://jwatte.com/blog/ai-quick-start-for-launching-a-business/
RSS: https://jwatte.com/feed.xml
JSON Feed: https://jwatte.com/feed.json
Hero image: https://jwatte.com/images/ai-quick-start-for-launching-a-business.webp