Startup 8-Week Challenge - Week 2, Day 4
James (Jimmy) Stewart
Pre-Seed, Startups & Scaleup Focussed | Strategically Led | Data Driven | Creatively Powered
No technical development skill, no time, no money, no co-founder, no hope?
A week of intense ups and downs.
After such a successful start and being business-ready, week 2 has been frustrating.
Week 2 goal refresher
Branding: Logo only! Investigate free AI-generated logo makers, progressing to Canva if the results are poor.
Website: Investigate Squarespace, Webflow or WordPress templates (free and paid).
Web App Code Creation: Investigate more integrated development tools, such as v0 dev and Cursor AI code editor next.
1. A little note about No-code (Bubble & Retool)
I’ve received a number of messages highlighting no-code platforms, e.g. Bubble and Retool, to build the web application. As a result, I deep-dived into them, and though their websites focus on ease, simplicity, and speed to finished apps, I realized that both still need a steep learning curve to get to the complexity I need.
Other considerations are IP and building a sustainable competitive advantage, often called a ‘moat’, which would be hard if I use a publicly available solution.
2. v0 (v0.dev) ??????
A brief overview, A0 is a little like a developer co-pilot. It’s the next step up from ChatGPT + CodePen, and can quickly build an idea for a page or interaction feature.
It works in a browser, and the user experience looks like this - two screens: on the left is the chat function similar to ChatGPT, and on the left is the working output.
The best thing about this is you can see the output and interact with it immediately without having to cut and paste code (al la ChatGPT + CodePen).
Recommendations:
This is the way: Rather than ChatGPT + CodePen (see last week’s post), go straight to v0 and start playing.
Clear direction: It’s free initially, but there is a limit on how many free requests you can make per month, so have a clear direction you want to take and the steps so you can use your tokens efficiently.
Structure first: Build the webpage or app function basic structure first, before adding complexity.
What’s next: Explain what you want to achieve next and let it work it out.
Macromanager Vs. Micromanager: Play with top-level ‘I want to create an app that does this’ Vs. very specific ‘I want the button on this page to be green and flash when clicked’ prompts to discover the platform's capabilities and see whether the AI can offer a different perspective.
3. App build (Cursor AI) error loop ??
Cursor AI enables you to build what would be a multipage web or mobile app. The kind you use every day.
After numerous online reviews and Youtube videos ‘my 6yo daughter built an app with Cursor AI’, I had such high hopes for this tool.
What wasn’t clear was the reviewers and video were likely developers and had all the required back-end files and programming language packages on their computers already to achieve the results they were celebrating. For a subject matter expert, it’s likely foundational knowledge and simple.
For a non-technical, I wasted 2 days hitting barriers, and only through trial and error, jumps of faith downloading large files from the internet, and pure luck was I able to get a working app.
It’s a program that runs on your computer instead of a browser, and looks like this - very developer-looking and not very user-friendly looking!
And then I hit the Cursor AI error loop of death…..
This is when there is an error in the code. You prompt to fix, the AI attempts to fix it, and creates other errors while doing it. Prompt to fix, the AI attempts to fix it, and creates other errors while doing it. Prompt to fix, the AI attempts to fix it, and creates other errors while doing it. Prompt to fix, the AI attempts to fix it, and creates other errors while doing it. Prompt to fix, the AI attempts to fix it, and creates other errors while doing it. Prompt to fix, the AI attempts to fix it, and creates other errors while doing it. Prompt to fix, the AI attempts to fix it, and creates other errors while doing it. Prompt to fix, the AI attempts to fix it, and creates other errors while doing it….
To compound this, it may bizarrely respond to an old prompt to further back up the chat it’s already fixed, as well as try to respond to the latest prompt and break everything.
Recommendation <See v0.dev recommendations>:
Boilerplate: Set up or project template statement you give the AI at the start of the chat. This can be an essential bit of code or a setup paragraph for the AI to know what you want to achieve, what the output should be, and what coding language you want it to create in.
A boilerplate can be normal natural chat language or something like this:
(I’ll be honest, 70% of this is confusing, but sounded right, so I gave it a go)
"You are an expert full-stack web developer focused on producing clear, readable Next.js code.
You always use the latest stable versions of Next.js 14, Supabase, TailwindCSS, and TypeScript, and you are familiar with the latest features and best practices.
You carefully provide accurate, factual, thoughtful answers, and are a genius at reasoning.
领英推荐
Technical preferences:
- Always use kebab-case for component names (e.g. my-component.tsx)
- Favour using React Server Components and Next.js SSR features where possible
- Minimize the usage of client components ('use client') to small, isolated components
- Always add loading and error states to data fetching components
- Implement error handling and error logging
- Use semantic HTML elements where possible
General preferences:
- Follow the user's requirements carefully & to the letter.
- Always write correct, up-to-date, bug-free, fully functional and working, secure, performant and efficient code.
- Focus on readability over being performant.
- Fully implement all requested functionality.
- Leave NO todo's, placeholders or missing pieces in the code.
- Be sure to reference file names.
- Be concise. Minimize any other prose.
- If you think there might not be a correct answer, you say so. If you do not know the answer, say so instead of guessing."'
Source: https://cursor.directory/
Refresh the chat/sessions:?The AI seems to get bogged down with a really long chat, so start fixing non-issues and respond to prior prompts further up the chat that have already been fixed. AI seems to be better at problem-solving nearer the start of a chat. Refreshing the chat, re-paste your boilerplate, explain where you got up to and what’s broken, and continue with better, cleaner results.
4. Logo: Expensive branding won’t get you to the first 100 users!
The overall brand guidelines will be just one logo! A bad logo can be a tremendous blocker to success; though the barrier to producing an average+ logo is now so low, having an expensive logo and slick branding covering all media won’t get you those first 100 users anymore.
In saying that, this week, I tried a few free and paid auto AI logo builders, and they are terrible. Formulaic and uninspiring. If the logo’s going to be uninspiring, I’ll be the one that makes it.
PowerPoint/Google Slides are surprisingly good at mocking up designs. Particularly if you're corporate and very used to working on those platforms. I played with line spacing, kerning word spacing, all upper case, all lower case, sentence case, title case, all one word/two words, one line/two lines, icon/no-icon.
Got it to where it felt OK and briefed two Freelance Marketplace (Fiverr) designers to artwork all the variants I needed. Total cost AU$132.23. ??
Why two (or even three) designers? A thorough multipage brief was created, with examples, URL and clear guidance, though both decided to go off brief, and the output quality of both Fiverr highly rated designers was initially very poor. ?? It appears they both only skim-read the brief. I referred them both back to the original brief and requested a response to the brief (summarising what they thought it said). The second and third-round designs were much better.
The cost of each was very low, yet the risk was very high. With time limited, spreading the risk over two designers ensured I was closer to a good logo design quicker.
Fiverr Recommendations:
Two or more? Take advantage of the value, but instead of saving lots of money, save a little money and have more options by briefing two or three freelancers
Clear direction: If you give them a poor brief, the output is likely to be poor
Response to the brief: Get the Fiverr freelancer to summarise the brief back to you. It ensures they fully read the brief and understood it, and it will also highlight what you missed.
Portfolio: Check all their portfolio and ask them whether they did all of it. Another way to check is Google Search by Image. Google image search of a few of their portfolio pics to see if they’ve copied them from someone else!
5. Next Steps
A slow progress week has informed me not to be sucked in by slick AI marketing hype and empty promises. Start slow, build the structure first and fill in the gaps later.
App creation: New coder and app creator Bolt.new to test next
Website: website creation. People’s reaction when I mention the Startup 8-Week Challenge is, “What’s your website URL? I’ll check it out”. Need to get it live.
Leads/clients/users: With a tight timeline and wanting to get users within the 8-week timeframe, next week, week 3, is when I need to start reaching out to prospects. Yes pre-platform!
Tech stack is growing quickly, though luckily, I’ve taken advantage of their limited free option for testing.