Kevin Michael Schindler

Want a self-improving AI Agent that actually takes user feedback? 🤖✨ Here’s how you can build one (step-by-step): 1️⃣ Create an AI Agent using your favorite platform. 2️⃣ Add a tool that can write to a txt-file, database, or JSON. (Doesn’t matter which—just make sure it stores feedback!) 3️⃣ Set up a prompt template that reads from the same file. This way, your agent can learn and adapt as feedback comes in. 4️⃣ Teach your agent to recognize user feedback/preferences and update the file. Example preferences: 👉 Always use emojis 😄 👉 Shorten lengthy replies ✂️ 👉 Stop yapping💀 5️⃣ Test it in real-time! Interact, give feedback, and watch the agent grow smarter with every interaction 🚀 6️⃣ Be impressed by how simple yet effective this process is 😍

We are automating things the most expensive way possible. I see this symptom everywhere right now: Skyrocketing software bills with no real increase in business output. Founders build an incredible workflow in an AI chat window. It works. It's smart. It feels like magic. But then they try to scale it exactly as it is. Let’s look at the reality of executing a repetitive digital task. The cost hierarchy is brutal: → 𝟭. 𝗗𝗶𝗿𝗲𝗰𝘁 𝗔𝗣𝗜 𝗥𝗲𝗾𝘂𝗲𝘀𝘁 (𝗡𝗼 𝗔𝗜): Fractions of a penny. Instant execution. → 𝟮. 𝗔𝗜 𝗧𝗿𝗶𝗴𝗴𝗲𝗿𝗶𝗻𝗴 𝗮𝗻 𝗔𝗣𝗜: A few cents. Efficient, but adds reasoning overhead. → 𝟯. 𝗔𝗜 𝘂𝘀𝗶𝗻𝗴 𝗠𝗖𝗣𝘀 (𝗠𝗼𝗱𝗲𝗹 𝗖𝗼𝗻𝘁𝗲𝘅𝘁 𝗣𝗿𝗼𝘁𝗼𝗰𝗼𝗹): Dollars. The model burns expensive tokens just trying to understand its environment and tools. → 𝟰. 𝗣𝘂𝗿𝗲 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁𝘀 (𝗖𝗵𝗮𝘁 𝗪𝗶𝗻𝗱𝗼𝘄): Financial bleed. Relying on an AI agent's "skills" in a chat window means you are paying a supercomputer to read, reason, and guess its way through a repetitive task from scratch, every single time. AI agents are incredible for discovery. If you want to move fast, test ideas, and figure out a workflow—use the chat window. Let the agent fumble through it. But the moment that workflow is proven? Stop relying on the agent to execute it. Instead, use the AI to write the code for a traditional, background automation. The cost drops by 99%. The reliability goes to 100%. Using a frontier LLM to run a static, repetitive task is like hiring a PhD to do manual data entry. It works, but it’s a massive waste of resources. Build workflows with agents. Scale them with scripts. Which processes in your business are still running on expensive AI agents that should just be automated in the background by now?

Good thing you do the planning. You can define budget or time constraints as you would do for yourself. "I want to implement a new feature and ship it to production in 2 days. Whatever we are implementing should not increase my software or computing costs by 200 Euro per month" So yes, the solution is: "just give it more context"

My biggest challenge with vibe coding / agentic engineering lately has been getting stuck in what I call a "plan doom loop" - have AI write a plan - review myself, seems good - have AI review plan, it always finds something - repeat It drains my time and energy to determine how important the "findings" really are Who has solved this

@NafetsWirth Hab den Ralph-Lauren-Loop-Trend leider komplett geskippt. Ist vllt das gleiche in grün xD

@KevinMSchindler neue ralph loop

I have a confession. I used to give terrible advice about AI. I told everyone to avoid multi-agent workflows. Honestly? I still think the whole "Agent Lisa talking to Agent Jeffry" setup is mostly a gimmick. It usually just burns tokens and wastes efficiency. My advice was always to simplify and optimize your single prompt. But after journeying through billions of tokens and building dozens of workflows, I realized something.... There is exactly one specific use case where a multi-agent setup is 100x better than a single prompt. Self-correction. Think about it: "You don’t have to be a chef to know the soup is salty." The same applies to LLMs. The AI generating your faulty output is actually smart enough to recognize its own mistakes. We are stuck in the habit of doing the heavy lifting ourselves. We get a bad output, and we spend hours manually tweaking the prompt to fix it. Instead, you can bootstrap and auto-improve the prompt by showing the output to the AI itself. → Agent 1 generates the draft. → Agent 2 critiques the draft and identifies the flaws. → The system automatically rewrites the prompt based on the feedback. Run this in a loop, and you have a self-optimizing system. It is time to start treating it like a system that can QA its own work. Where are you still manually fixing AI outputs when you could be automating the QA?

@zekramu Wasn't lite-llm the package that recently got compromised?

it is cheap. so fucking cheap. $40 for the amount of tokens you get is actually criminal. proxy the inference harness via litellm and burn through them shits. you can literally hook it up to Claude Code.

@itsolelehmann Guten Flug!

Cyprus 🇨🇾 -> Berlin 🇩🇪 Your luggage goes 3x with a baby LMAO

@NafetsWirth was moinsch?

@KevinMSchindler Kein Loop

@collin_ruth89 ~500 bucks...for anthropic

You created 7 AI agents last month? Cool. How much revenue did they generate you?

In the coding world: - open 4 terminals - claude --dangerously-skip-permissions - Add Claude API key - give access to .env with credentials - "Don't ask for feedback."

@bitcoinpanda69 How old are you, foo? I think for some gens the problems are clearly "money" and "job"-focused. Past that there are other problems, but I think they are downstream from this one.

My strongest blackpill belief is that UBI will lead to an overwhelming amount of mental illness most people if they are honest w themselves need external boundaries and challenges imposed on them You're asking everyone on the planet to self-govern their whole lives when most cant even go to bed on time now when they have work early

@BridgetMWillard If you haven't tried already, I highly recommend giving it one try. The easiest is to just activate the MCP/Connector in Claude and add a wordpress skill from here: skills.sh/?q=wordpress The first time, connecting Claude to a SaaS application felt magical for me :D

@KevinMSchindler Okay there you go. Love the use case!

Is anyone in WordPress doing WordPressy things lately? All I see is Claude This Perplexity That.

@BenjaminHouy "Made in Germany" doesn't feel the same anymore. Now it just refers to where it is made.

It’s precisely this arrogance that made me leave Germany. There is this sense of superiority that permeates through so much of German society. Many Germans simply seem incapable of accepting criticism and acknowledging that there is room for improvement. Sadly, an arrogant society is a doomed society,

Yes, kind of. However, Claude can then also plan and do everything on WordPress as well via the MCP. I find the switching between tools annoying. I never realized how much "mental" power it drains to constantly "search" for things. I had a great experience moving everything into a command line.

@KevinMSchindler Okay. So you like just talking to it. So basically Claude has replaced the rubber duck for programmers.

Python + Claude = Your new PC superpower. 🐍 I’ve been experimenting with creating custom widgets that interact directly with my computer. It’s fast, lightweight, and honestly, a lot of fun. If you’re looking for a way to make your PC work for you, this is it.

@zarazhangrui Hire someone so they can enter the matrix for you xD

My single biggest pain point right now: AI-induced attention deficit I have 5 Claude Code sessions running 10 Terminal tabs open 50 browser tabs open 100 X articles in my bookmarks When there are multiple AIs working for you, you're constantly task switching When you're waiting for AI to output, you're constantly distracted Because what used to be hard is now so easy, you always wanna do more, more, more Because of the speed of the AI news cycle, you feel a constant need to be on this app Add all this up and the result: you live in a constant state of distraction & delirium. Focus & deep work is harder than ever If anyone has figured this out pls teach me how

@benjaminprinter Good reminder

Every mf with a case study could sign a $50K deal if they reframed it from being a ‘success story’ Buyers in B2B don’t give a shit about upside When you say: “We helped Client go from X and grow their revenue by 42% in 6 months” The buyer is forced to imagine their current situation and change it to a better future on the basis of a possibility The problem is they aren’t currently experiencing it So it’s just a vague claim to consider Which makes it optional So they can appreciate it without needing to pay you because nothing about their current reality is forcing them to move forward In B2B environments there are always careers on the line and political decisions being made No one gets fired for missing a ‘growth opportunity’ But people definitely get fired for missing something that could have been prevented So if your case studies are framed around growth, they’re still optional But when you simply reframe them around the angle of ‘We helped the exec team avoid a $4.8M blind spot that they hadn’t realised in 36 months” They can defend your offer This is a formula you can apply to every single case study that you have Because the general formula is - Before - After - Result When you can simply reframe the exact same thing to - Hidden Problem - Escalation - Near Failure - Intervention - Stabilization So an example of this is: “Mr gayboy client was spending $85K/month on paid channels and thought performance was ‘fine.’ But their funnel was under converting by 30% relative to benchmark, which meant 140 qualified buyers per month were being lost silently. Over a 5-month period, that translated to roughly $3.5M in missed pipeline. Which meant the issue was simply how demand was being processed post-click Once we rebuilt the conversion layer, we tightened up the $3.5M bleed within 90 days.” So you’re entire proof becomes the fact that you see what others are missing instead of the fact that you can produce upside Because the fastest way to prove you know what you’re talking about in B2B is always to expose a hidden lose that feels uncomfortable when its revealed The buyer immediately identifies it because it’s the lie they’ve been avoiding for years And it becomes the purest driver of urgency

@explorersofai Give up "everything". ...50% of this is worth nothing. Most people are just automating annoying tasks, that's all.

I’m so tired of hearing “Claude this, Claude that.” You’re giving one company power over all your sh*t. It’s the stupidest thing I’ve ever heard humans do. I can’t believe how quickly people are willing to just give up everything.

@NafetsWirth @d4m1n @alberduris /superpowers Figure it out (mega think pls)

@d4m1n @alberduris let the agent figure it out

why isn't OpenAI just spinning up 8000 parallel agents then?