JoJo 🔨 LegalCents

I'm a Principal Data Engineer at a big corporate. But for the past 12 months, I've been secretly building a SaaS on the side. It's now profitable. Here's what I learned building LegalCents while keeping the day job 🧵

@dhh Constraints breeding creativity is the one that resonates most. Building a SaaS with 6 hours a day forces you to cut everything that isn't essential. No time for over-engineering, no time for features nobody asked for. The constraint is the strategy.

What a great conversation about liquid v crystallized intelligence, constraints breeding creativity, being wrong about Facebook, taste v data, the people I admire, and what worked for us that probably no longer would for others. Thanks Sam and Shaan!

@sobedominik Same shift here. I barely open my IDE anymore. Terminal with Claude Code for building, browser for testing. The IDE feels like overhead now. The only thing keeping me in it is git diff views — but even that's getting replaced.

What's your coding stack these days? I feel like with agentic coding it becomes less and less important to be in the IDE for the majority of the time but I'm still defaulting to Cursor because I can use: – Terminal with Claude Code – Git viewer – Quickly edit code I've seen people use things like Cmux more and more and just curious what your stack is these days?

@code_rams This is the right pattern. Instead of the agent guessing how your infra works, you feed it the context directly. We're going to see every major platform ship a plugin like this. The agents that win will be the ones with the best context, not the best models.

Vercel just shipped a plugin for coding agents. It watches your file edits and terminal commands in real time, then injects the right Vercel knowledge into the agent's context automatically. What's included: 47+ skills (Next.js, AI SDK, Functions, Storage, Turborepo) 3 specialist agents: AI Architect, Deployment Expert, Performance Optimizer 5 slash commands: /deploy, /env, /status, /bootstrap, /marketplace. Live validation that catches deprecated APIs as you build No setup. No prompting. Skills fire based on what the agent is actually working on. Both my products are on Vercel. My coding agent (Vasi) handles deploys, env vars, and debugging. With this plugin, it gets Vercel-specific expertise injected automatically based on what it's actually working on. That's the difference between a generic agent and one that knows your stack.

@forgebitz The position tracking is the feature I didn't know I needed. Most SEO tools show you traffic. Knowing where you actually sit in AI search results is a completely different signal. This is where discovery is heading.

had some fun with this new position chart you can now quickly track: - visibility - mentions - position

@tdinh_me Just tried GPT-5 Nano for a small AI feature in my product. Not as fast as the benchmarks claim, at least for my use case. Still shipping with it though — good enough today beats perfect next month.

GPT 5.4 Mini and Nano available now on typingmind.com 😊

@forgebitz When the API goes down your users blame your product, not the model provider. Had to build fallback logic and a queue system just to handle the reliability gap. It's a real tax on small teams.

with the amount of outages anthropic is having it's almost impossible to put it into consumer/end-user products

@kepano "Runs locally. Just rules, no AI." — this is such a confident product statement. Not everything needs a model behind it. Some problems just need clean rules and good design. Been waiting for this.

I have been working on Obsidian Reader for a over a year. I didn't want to share it until I felt it was good enough. It's finally there. Consistent formatting for any article. Outline, syntax highlighting, nice footnotes, adjustable typography. Runs locally. Just rules, no AI.

@asmartbear This changed how I approach demos. I stopped trying to create urgency and started asking "what does your current process look like?" The moment they describe the pain themselves, the product sells itself.

The sales pitch is showing the customer that this product fits their needs and constraints. If you find yourself convincing them they need to buy something, or that their needs are stronger and more acute than they thought, that’s a waste of time.

The whole system runs on a spreadsheet and one hour of weekend prep. Everything stays tracked, nothing falls through the cracks. And I'm already working on automating more of it. If you're a technical founder selling to a market you're not part of — find someone who is. One person with the right network changes everything.

I don't have a network in legal ops. Never worked in legal. Don't know any GCs personally. But my SaaS sells to in-house legal teams. Here's how I make that work.

My co-founder is an in-house legal counsel. She lives in the world our product serves. Each day she picks 2-3 people from the spreadsheet, takes the draft I wrote, and polishes it with her own experience — finding something real that connects her to that person. It takes her a few minutes. But the reply rate is insane compared to when I was doing outreach myself. Because it's not a pitch. It's a peer reaching out.

Every Sunday after my son falls asleep, I spend about an hour finding 15 senior legal counsels, heads of legal ops, and GCs on LinkedIn. I run an automated process to pull their job titles, profiles, and descriptions — then generate a custom draft message for each one and put it all in a spreadsheet. By Sunday night, the whole week's outreach is ready to go.

@olamide_ai_dev Exactly. Same product, completely different conversation depending on who's across the table.

@jojo_data People you pitch to will determine how your service will be valued.

Closed a deal last week at 30% off our asking price. Doesn't feel like a win. But a month ago this same deal was headed for 50% off or dead. My co-founder went above the contact to his boss and reset the conversation entirely. Sometimes the best sales move isn't a better pitch — it's finding the right person to pitch to.

@forgebitz I've been using GPT-5 Nano for a small AI feature in my product and it's way slower than it claims.

4x more expensive somewhat similar to the new gemini flash lite a bit smaller and a lot more expensive

@kepano This is why I keep everything in markdown. The moment your data is locked behind a proprietary format or a company's decision to pivot, you lose. Files you control will always outlast the app that created them.

don't make me tap the sign

@marclou Did this exact test. My co-founder showed our landing page to three lawyers. Two of them didn't understand what it did. We rewrote the hero that week and demo requests went up immediately. Simple language beats clever copy every time.

The Girlfriend Test: Show her your landing page. If she doesn’t get it in 4 seconds, rewrite the hero.

@forgebitz The gap between what these dashboards show and what's actually happening is insane. 0% visibility on paper while ChatGPT is citing you half a million times. Feels like the tooling hasn't caught up to how discovery actually works now.

the problem with all these vibe coded prompt trackers is that they don't work at all 0% visibility according to made up metrics half a million citations from chatgpt in reality i understand why SEOs are having a mental breakdown thanks to this slop

@tdinh_me This is so true. The indie hacker space is one of the few places where someone building something similar to you will actually share what's working for them. No zero-sum thinking. The market is big enough.

Funded start-ups building the same product become competitors. Indie start-ups building the same product become friends. Indie start-ups only need $10k MRR to succeed. The market is so big that it is unrealistic and unnecessary for two independent developers to “compete” with each other. Both can support each other and win together. The only rule applies: have your own touch and vision for your product. Don't be that annoying copycat. Build wealth on your own effort and talent.

@gregisenberg Claude Code has changed how I build. I used to context-switch between 5 tabs. Now I describe what I want, review the output, and ship. Went from spending weekends debugging to actually building features.

claude cowork and manus ai are probably two of the most underrated ai tools I can think of

@asmartbear The retention problem is real. People love the idea of AI features but churn when it doesn't actually solve their specific workflow. The ones that stick are the ones that quietly make one boring task disappear — not necessarily the ones that lead with "powered by AI."

No one cares whether it was "made with AI," but many people are interested in trying things that allow 𝘵𝘩𝘦𝘮 to use AI. However the key word is "try." Cancellation can be 90% at 4 months. It has to also work, which it typically doesn't (currently).