
Service and CRM CTO @IBM
Atlanta, GA
Joined September 2009
- Tweets7.7K
- Following1K
- Followers6.9K
- Likes4.4K
You ask an AI agent for something big, it says done, and the build does not even succeed. Then each try costs another 30 minutes of waiting for the automatic checks.
That was the situation for David Cortés on Shopify's Polaris team. The idea that helped came from autoresearch, a small GitHub project by Andrej Karpathy: give the agent one number to improve, let it try again and again, keep what runs faster and throw away what crashes or runs slower.
In my own reading, in David's case the agent had a clear goal and a clear way to measure it.
How long does one try take in your own checks, and could a script measure it?
#AIAgents #Autoresearch #SoftwareEngineering
youtu.be/j2yX6Y9vhJo
1
2
402
If an AI agent gave you an answer today, could you explain it without the agent's help? This is becoming trend to trust LLM agents answer and output blindly impacting Critical thinking capabilities of humans.
There are six questions to ask. Three questions come before the agent answers:
- Did you form a hypothesis? Even a bad guess helps, because a polished answer makes other ideas hard to imagine
- Are you asking it to help you think, or to think for you?
- Would you bet on this in front of the person who must maintain it?
Three questions come after:
- What would have to be true for this to be wrong? Go and test it
- Have you heard the counter argument?
- What are you no longer practising?
Which of the six do you skip most often?
#CriticalThinking #AIAgents #ClaudeCode
youtu.be/T1HorOZ_cZ4
287
AI is changing how we work in the Salesforce ecosystem, and I believe every role will need to evolve with it.
Whether you're an Admin, Developer, Business Analyst, Consultant, or Architect, the expectations are changing.
But one thing hasn't changed: Strong fundamentals still matter. In fact, they matter more than ever.
I've put together a YouTube playlist covering different Salesforce roles, what they actually do, and how they can adapt and grow in the age of AI.
youtube.com/playlist?list=PL…
I'm curious to hear from You : Which role do you think AI will change the most, and why?
#Salesforce #Agentforce #AI #SalesforceCareers
2
8
923
One task is limited by the model. Many tasks are limited by you.
The Super-agent, an agent design from the DeepSeek Harness team, starts from that limit. In many agent designs, each piece of work lives in its own session, a separate conversation with its own memory. You switch between them by hand and remember what each one holds. The DeepSeek Harness team says that in their experience a person can usually follow only 3 to 5 active sessions at once.
Their answer is to work with one continuous agent right where you already work, such as team chat or GitHub. A Router handles the people, and Workers do the tasks.
Connecting chat or email to a single agent is not enough, the team says. It still waits on long jobs, tasks get mixed together, and everything stops when your machine goes offline.
How many AI sessions can you keep track of before something slips?
#AIAgents #AgentDesign
youtu.be/JLc8qCtTGwU
2
2
332
I've seen countless jokes about AI-generated code and vibe coding. Never imagined I'd witness something like this within my own network!
But behind the humor is an important lesson.
In the age of AI and vibe coding, strong fundamentals matter more than ever.
AI has dramatically lowered the barrier to building things. What once took days can now be accomplished in hours, sometimes minutes.
But generating code is not the same as engineering a solution.
Understanding architecture, security, scalability, design principles, and engineering best practices is what separates a quick prototype from a reliable, production-ready solution.
And that's where the real opportunity lies!
AI doesn't make foundational knowledge irrelevant. It makes that knowledge exponentially more powerful.
My advice to everyone embracing AI:
Strengthen your fundamentals. Understand the why, not just the how.
Embrace AI wholeheartedly. Use it to accelerate learning, experimentation, and innovation.
Think critically. Never confuse AI-generated output with a validated solution.
Follow engineering best practices. Security, testing, scalability, and maintainability still matter.
Stay curious and keep learning. The tools will evolve, but your ability to reason, question, and solve problems will remain invaluable.
The future isn't about choosing between AI and human expertise. It's about combining both to achieve what was previously impossible.
Don't just learn how to prompt AI. Learn enough to challenge it, guide it, and make the most of it.
The stronger your foundation, the higher AI can help you build.
#AI #VibeCoding #SoftwareEngineering #ContinuousLearning #EngineeringExcellence
Salesforce's Braze announcement does not explain how the products will connect.
So I read the product pages and wrote down my own reading of the fit.
- Data 360 turns a company's data into business context for agents and teams
- Braze turns live customer signals into messages and journeys
- Marketing Cloud runs campaigns, and Salesforce expects Braze to strengthen its cross-channel personalization, with deep mobile and in-app experiences
- Agentforce, Salesforce's platform for AI agents that answer questions and take actions, sits beside BrazeAI
That is a reading, not a plan the companies have shown. Bill Magnuson, Braze's chief executive, said Braze will keep building as Braze.
How do you read a fit like this before the vendors explain it?
#Salesforce #Braze #MarketingCloud
youtu.be/b9SvB1-Ce5E
Salesforce calls admins the backbone of every org. Which explains why everyone leans on you.
What admins do every day, which tools to learn first, and where #AI fits in.
#Salesforce #SalesforceAdmin
1
159
Saving one record in our app took about seven seconds (our estimate). Long enough to click Save again, just to check.
The records sat in Delta tables behind a SQL warehouse. Here is why we moved them to PostgreSQL.
jitendrazaa.com/blog/sql/dat…
#Databricks #PostgreSQL #SQL #Performance #Architect
youtu.be/5kqriNmG03w
1
181
Should an LLM control a 6G network?
Our IEEE VTC2026-Fall paper says: not on its own.
The LLM turns an operators goal into structured objectives. Guardrails check them, and deterministic controllers run only what passes.
In our 260-intent benchmark that cut the unsafe accept rate from 100% to 0.8%, with a 4.6% false-reject rate and a 50% escalation rate.
Paper: Safe Intent Compilation for 6G Network Orchestration Using Guardrailed LLMs, with @amit_sfdc . #6G #LLM #IEEE
116
Jitendra Zaa retweeted
A guy in India bought a Sony Bravia TV. It developed green lines under warranty. Sony's service team called it "external damage" and refused replacement.
So he registered SonyBravia.com and posted the entire saga — emails, service records, photos.
Sony's response? A 6-page legal notice demanding he delete everything, apologize in 7 days, or face a ₹5 crore civil + criminal claim.
Indian Gen Z really isn't sparing billion-dollar brands. 😭
Jitendra Zaa retweeted
Meta Wanted To Replace Engineers With AI Agents - Literally Everything Went Wrong
- Internal docs show a 60% planned cut
- AI was supposed to do most work
- Zuck changed his mind, last minute
- Only did a 10% lay off
- Employees staged a huge revolt
- AI slop generated 40% more security incidents
The most surprising part of this story is assuming that humans wouldn't revolt 😲















