As AI moves from the research lab into production systems, there are new problems. E.g., the model had great scores when it was developed, but now it’s nearly useless. You probably need automatic data cleaning in your pipelines… 💧🛁🧴 #AITechniques #DataCleaning #CienAI #RobertKall

As someone who has done the jumps, there is a logical endpoint. If you can sell a $1 favorable outcome for $0.10, you have a rocket ship business… If that costs you $0.02 to deliver, you have the next unicorn (or better). #AI #BusinessModel #RobertKall #CienAI

There are terms like MLOps and Data Observability that people throw around. There are new vendors selling you expensive monitor solutions for your AI systems. All good. But a lot just comes down to common sense. Your AI is constantly being “tested” by new “bad” data, which is different from what it has seen before. It may not handle it wel,l and then your users suffer. #AI #MLOps #Pipelines #RobertKall #CienAI

A startup is a mission. A mission in a jungle, where you will constantly encounter new dangers. Your job as a founder is to try to anticipate most dangers, move the team safely to the next camp site, and if you’re face to face w/ a wild beast, either fight it or escape it. That's never easy, but it is doable. A good outcome is when you reach base camp. Then you can decide if you want to climb the mountain too (likely to be snow leopards on that trail…). #Startups #Mission #RobertKall

Spending a lot of time discussing and testing new UX paradigms. It’s clear that the chat interface is very useful, but companies like Anthropic have released tools like “Claude Artifacts” that let you generate code for interactive widgets on the fly. I think that’s fly… #AI #CodeGen #AgenticAI #CienAI #RobertKall 🪰

I love to get a historical perspective on things. Here is my take on the change that is happening right now and how it relates to past periods. And while the robot does the work in the next phase, remember that there is a lot of work to be done to tell the robots what to do in front of us. And that work is infinite (i.e., as we get better at telling robots what to do, we will think of more things to tell them). I am noticing this right now as I do a lot of “vibe-coding”. A feature that would have been impossible or just super long to implement is done in 2 hours, but it takes effort to define how it should work. #AI #FutureofWork #RobertKall #CienAI

No developer has ever inherited a system that is “perfectly designed”. And sometimes, inexperienced ones recommend the “Complete Rewrite” option. That has always been a recipe for disaster in the past, but now with Claude Code and CODEX, it may seem more appealing. Don’t do it! Instead, opt for splitting out the old system into X number of sub-components with the exact same functionality and legacy code base. Then use AI coding to rewrite them one (or two) at a time. You will limit the “blast area,” and you can move to a more modern stack and architecture, while keeping your users happy. #completerewrite #claudepilled #robertkall #cienai

Entrepreneurship continues to be glamourized, as something where you are in total control of your destiny, and working on creating wealth for you and your close ones. The reality is different. Most days have some kind of dread. Customers, employees, suppliers, etc. And unlike in a, there is no one coming to save you. But once in a while, you get that super high (big deal, big launch, great customer testimonial) that recharges you for the next few weeks of misery... # entrepreneurship #robertkall #cienai

I am back, spending almost 50% of my time on product, like many other "Claude Pilled" (although I prefer CODEX) technical founders, who all of a sudden noticed that they could shape their products directly again. Cool. Built so much in the last month that it will have huge benefits, but it comes at a cost (besides the tokens). There is just one more improvement, one more bug fix, one more architectural refactor that can be done. With the team, you often deferred the work. If you know you can do them yourself, you just push ahead in a world of coding abundance... #vibecoding #claudepilled #robertkall #cienai

Nothing is more exciting than a nice doomsday story. And you’ll get them daily related to AI through the news. It’s true that AI challenges existing business models and that employment will shift. But after working with Agentic AI for a year, I think the demand for people to create a new kind of company is enormous. These companies “close the loop,” so they actually know if their customers benefit from their products. Then they use AI to deliver better outcomes and more personalized services. The demand for outcome-based services (priced at a fraction of the value they deliver) will be infinite, and so will the need for workers who can create them. #AI #FutureofWork #RobertKall #CienAI

AI is non-deterministic, but that does not give you a free pass to deliver non-working agents to your clients. Add'l challenges involve context changes, new models, and modified instructions. How do you solve? Using something called “Evals” (short for evaluations), where you test each agent using another LLM to see if the responses it gives are correct or not. You can run those continuously and get numeric scores for each test, so if something changes, you’ll know immediately. #AI #Evals #Agentic #RobertKall #CienAI

As an entrepreneur, you always have to fight to gain people’s trust. This is often surprising to folks who have joined my startups after working in larger companies. Why do customers not just buy our products? After all, they are so much better than the incumbents’ alternatives. It all comes down to trust. BigCo has it. Your startup has to earn it. The good news is that it’s not hard to build trust. (at least in theory…) 🤝 #Entrepreneurship #Trust #CienAI #RobertKall

Despite massive investments in Gen AI, business value to date has been limited. Companies run on numbers, where Traditional AI is far superior. But Gen AI can generate new numbers (from unstructured data), for businesses to benefit from. 🔢 #AI #Analytics #GenAI #TradAI #CienAI #RobertKall

he is not wrong

What’s the big deal about Agentic AI? Here are 3 things that distinguishes it from current Gen and Traditional AI. It is ready to work for you! #AI #AgenticAI #CienAI #RobertKall

After 20 years as an entrepreneur, where sometimes I went several years with sleepless nights and “that sinking feeling”, I have discovered a fantastic fix: Just get up and do something every morning. 🏃‍➡️ # Entrepreneurship #ActionCreatesTraction #CienAI #RobertKall

I work a lot w/ propensity models (i.e., what is the probability of a specific outcome, such as winning a deal). The output is a percentage. So, how do you know if your model is any good? The deal did not close even if was a 70% probability of a winning deal… The key is to measure model quality. The AUC (Area Under Curve) is one useful measure. This visual explains the different levels of precision for different types of phenomena. Even if AUC is not 1, predictions can be very useful. #predictions #AIquality #RobertKall #CienAI

AI is inherently non-deterministic. Businesses want deterministic analytics tools that show the same numbers as yesterday. If you use AI to generate insights, remember that! 🦘 #AI #Analytics #NumberStability #CienAI #RobertKall

Startups is the process of taking an idea from nothing to something. The process is ugly and you are constantly thinking about all the things that are still not working. Until they do… 🔧 # Entrepreneurship #Mindset #CienAI #RobertKall