The Ledger
Every statement that passed quotation and attribution checks. Mix any filter with any other: certainty 1/5, debate potential 5/5, or both at once.
why aren't all 25 resolved? a statement only gets an assessment when the public record can support or contradict it. opinions and what-ifs never can, and 0 checkable ones are still open, waiting for their date. predictions held up or didn't; assertions are supported or contradicted. on every card: ▮▮▮▮▮ certainty · ▮▮▮▮▮ debate potential. speakers are clickable
Howell: Hardcode Proven Workflows With Scripts to Eliminate LLM Black Box Risk
“Once you've got a workflow that's working for you want to get rid of all the ambiguity. So I could have GrokBot write the blurbs every week and it does a really good job. But if the model changes or I don't know, just depending on how the bot's feeling, it mig…”
Howell: 5 Targeted Agent Pitches Outperform Mass-Emailing 200 Leads
“What you should do is set up an automation that goes on Monday that goes, okay, your job is to Add five prospects to our list. And then on Monday, okay, pick the top three prospects and create an ad package custom to them and draft an outbound email and then s…”
Howell: Build Physical Brands as Digital Creation Costs Approach Zero
“The cost of creating digital things is going to zero. There's no friction there. Actually making something physical a brand that you can just like build over time. I think that, I think that's a smart thing to do.”
Howell: Visual Agent Avatars Reduce Cognitive Switching Costs Over Text Interfaces
“Color and shape is all the human brain needs to differentiate something. So you can easily recognize all your different little teammates here versus like a grade, just like a gray thread in chat GPT or Claude. It's just, I find this tool saves you so much on s…”
Howell: Running multiple businesses on one GrokBot instance degrades agent effectiveness
“You don't have enough tokens to build four businesses as one at once. You're going to go into debt paying for tokens on even worse than that. You're going to have context bleed. That's going to make all of your agents less effective.”
Howell: Human operators must make tool choices instead of letting agents decide
“One of the pitfalls I see people make is having their agents make too many decisions for them. So for example, I've been struggling with where does our content live when we're writing the newsletter? Does it live in Notion database? Does it live in local files…”
Howell: Have chief of staff agent execute a workflow before building a dedicated bot
“Say you want to set up an outbound sales agent for your newsletter and sell ad space. Have your chief of staff do that exact task once, and then once it does it, and you've reviewed it, and it worked, be like, okay, now create an outbound bot that does that th…”
Howell: Cap AI Agent Status Reports at Five Lines to Save Tokens
“I think that you just five lines is all you need more than that. It's you're just going to be burning tokens in context.”
Howell: Three Rounds of Adversarial AI Review Improves Output to 90%
“Create a panel of sub agents for Grokbot. It could be, you know, the chief of staff or other agents on the team and have them do an adversarial review of your work. Do that in three rounds. And that has got, that has made me so much progress on stuff that norm…”
Howell: Operators must spend weeks in the weeds before agents run autonomously
“You really have to run, you have to be in the weeds with the team for a couple of days, a couple of weeks before you can run it autonomously.”
Howell: Agent Creep and Unnecessary Bots Burn Tokens Without Making Progress
“Why I like rock bot and how I use it is I only add an agent if we really like needed, I just, I'm so anti idea creep and agent creep because that's how you burn tokens. That's how you'd spin your wheels. That's how you don't make process progress.”
Howell: Over-Engineering Agent File Sync Workflows Distracts from Actual Output
“Grokbot works well on its computer and it can read stuff on yours, but I find myself falling into this trap of, okay, you know, generate a report and then put it on my desktop and then put it in notion. And I don't know if other people are like this, but you e…”
Howell: Never Let ChatGPT Specify Tech Stacks for Replit Agent
“I think it just will cross too many wires if ChatGPT is like, use like Next.js and Node and this and that, because that's what Replit's good at, is kind of picking the stack that it knows how to do, so.”
Howell: Including OAuth in Specs Confuses AI Coding Agents
“If it, if ChatGPT tells you OAuth, like, that's probably going to confuse your AI agent. It's just more steps.”
Howell: Build 50 Apps and Iterate Constantly to Master AI Development
“I would say go make, like, 50 apps. Just, like, keep iterating. Don't get stuck on one. Like, that's my advice is to If you have an idea, try to make it three different ways and then move on to the next idea and you'll get really good at using all these tools …”
Howell: Find a buyer before building software rather than after
“That's where most people get this wrong is they build something and then try to sell it or try to get users. But I found it's a lot easier to just find something someone wants to buy and then build it.”
Howell: Replacing existing software makes custom app sales easier
“But I think it's an easier sell when you're replacing something or doing something better than like existing software could do.”
Howell: CRUD apps are the sweet spot for AI vibe coding
“Usually we're looking for crud apps, right? Something where it's just putting data in a database, pulling it, visualizing it. You could do all of this with vibe coding for sure. It's just going to take, you know, more tinkering.”
Howell: Most automations can be turned into front-end apps
“Automations is also a good keyword to look for because you can turn most op automations into an app with a front end.”
Howell: AI coding tools consistently struggle with dates and time zones
“AI in general gets kind of mixed up with date formatting and time zones. So that's something to look out for sure.”
Howell: Paying $50 for debugging bounties beats hours of AI troubleshooting
“It's just like, 50 bucks to troubleshoot something is such a good use of money versus me banging my head against the wall, burning 10 dollars in AI credits and six hours of my night, you know?”
Howell: Pitch custom AI apps to replace clients' most expensive SaaS
“So I think what you can do if you're like an entrepreneur trying to do this is like, Build like one or two apps, like probably more than that, like get good at it. Like I made like 40. And then just call business owners and be like, what's the most annoying th…”
Howell: Letting Users Paste Invoice Links Reduces Onboarding Friction
“And that also, that features a good way, I think, to get users who, you know, may not want to switch over their whole billing to you, right? They just want to use this to push existing invoices.”
Howell: Sending Loom demo prototypes wins $2,500 Upwork gigs
“If it's a simple enough app, and it's 2500 bucks, right you could go ahead and build the prototype, and then in the proposal, send it as a Loom link of you, like, demoing it for them, which I've done, and it worked for the first one I ever did.”