|
Your Weekly AI Briefing for Leaders
Welcome to this week’s AI Tech Circle briefing, clear insights on Generative AI that actually matter.
While writing any post nowadays, I always think, 'Is it worth writing?'
When all the tips, knowledge, and how-tos are available with the touch of a button via any LLM app, and, frankly, I am still thinking: should I keep curating and sharing this or just stop? This also reminded me that my first public post on any tech topic was back on August 20th, 2011, "Oracle Fusion Applications Foundation – Functional Setup Manager."
Today at a Glance:
- When GenAI is not to be used
- AI Weekly news and updates covering newly released LLMs
- Courses and events to attend
|
|
|
Claude Cowork
Anthropic is experimenting with something new called Claude Cowork; it's moving AI closer to execution rather than just conversation.
Cowork allows Claude to work directly with files and tasks, reading, organizing, drafting, and completing multi-step work inside a user’s environment. The shift isn’t about speed or automation hype; it’s about where AI sits in the workflow.
And the whole game is how easy it is to bring Gen AI to the existing workflows
Why It Matters:
Most AI tools still add steps: prompt, copy, paste, verify, repeat. Interestingly, Cowork removes those steps by letting AI carry work forward once direction is given. That changes productivity from faster responses to fewer handoffs.
Claude Cowork is an early signal of how AI becomes useful, quietly, by disappearing into the workflow.
|
|
|
When GenAI Should NOT Be Used
(And Why Saying “No” Is a Leadership Skill, and it's hard to follow during the last 3 years)
Over the past three years, I’ve noticed something interesting in GenAI discussions, and this also reminded me of the situation we were in during the Blockchain boom, when I wrote: "Blockchain is a Solution for Every Business Problem." Now, once again, not every use case requires Generative AI.
Everyone, including myself, is excited to say, “Yes, we can use AI here.”
Very few people are comfortable saying “No, and here’s why.”
From experience, that hesitation is costly.
1: When the Problem Is Still Not Clearly Defined
I’ve been in sessions where teams say:
“Let’s use GenAI to improve efficiency.”
- Efficiency of what?
- Speed of which decision?
- Improvement for whom?
If the problem statement keeps changing, GenAI will only amplify the confusion.
What I do now: If the problem can’t be written in one clear sentence, we need to pause the AI discussion and identify the problem.
2: When Accuracy Is Non-Negotiable, and Verification Is Weak
There are environments where:
- A wrong answer is not “acceptable with correction.” It is damaging by default
Examples I’ve seen:
- regulatory interpretations
- financial approvals
- legal conclusions
- compliance certifications
If there is no clear verification or review loop, GenAI introduces risk faster than value.
What I do now: We should only allow GenAI to assist in these contexts, never to decide or finalize.
3: When the Real Issue Is Organizational, Not Technical
I’ve seen teams reach for GenAI to fix:
- broken processes
- unclear ownership
- slow approvals
- misaligned incentives
AI doesn’t fix organizational dysfunction. It automates it.
What I do now: If a process is broken manually, we fix that first, then consider AI.
What Saying “No” Has Taught Me
Success often comes from:
- fewer, better-chosen use cases
- slower but deliberate adoption
- firm boundaries around where AI shouldn’t go
The Question I Now Ask First
Before recommending GenAI, I ask myself:
If this system produces a wrong answer, Will we notice quickly and know what to do next? If the answer is unclear, GenAI is not the solution yet.
|
|
|
Top Stories of the Week: Bifurcation in Enterprise AI Maturity: Organizations are splitting into two camps:
"Winners" who are rearchitecting their organizations for AI, and "tourists" stuck in endless demos
"Winners" are redesigning data architectures, processes, governance, and operating models to scale generative AI effectively, treating it as foundational infrastructure rather than a bolt-on tool.
Tourists remain stuck in pilots due to inadequate foundational changes, limiting ROI.
For leadership, this means prioritizing systemic overhauls, focusing on trust, security, and workflow integration, to avoid stagnation and capture productivity gains.
|
|
|
Favorite Tip Of The Week:
The Modern Software Developer taught at Stanford University, Fall 2025.
This course covers how coding LLMs are transforming every stage of the software development life cycle. The assignments are intended to take you from noob to expert in how to use AI to improve your software engineering productivity.
|
|
|
Things to Know...
The GenAI Mistakes I See Repeated in Enterprises
One thing I’ve learned over the past year is this:
GenAI initiatives don’t usually fail loudly. They fail quietly, without notice, by delivering less value than promised compared to when the initiatives started. Mistake #1: Starting with the Model Instead of the Problem
I’ve seen teams spend weeks debating, including myself:
- GPT vs Claude
- fine-tuning vs prompting
- token limits
Before anyone could clearly answer:
“What decision or workflow will this improve?” When the problem is not identified, the solution will always disappoint, and the issues remain the same.
Mistake #2: Treating GenAI Like a Feature, Not a System
Many pilots look impressive in demos, and we are all getting masterful in it :)
- clean UI
- fast responses
- confident answers
But behind the scenes:
- no data ownership
- no update process
- no failure handling
Mistake #3: Overloading AI with Too Much Context
This one surprised me early on. Teams assume:
More data ... better answers In practice, I’ve seen the opposite:
- diluted responses
- inconsistent reasoning
- loss of relevance
What This Changed for Me
These experiences changed how I think and evaluate GenAI initiatives.
Today, I’m less impressed by
- demos
- benchmarks
- early excitement
And more interested in:
- decision clarity
- operational ownership
- governance maturity
- long-term trust
|
|
|
One Action for CIOs: Make enterprise data AI-ready before making teams AI-faster.
Choose one critical domain (finance, supply chain, customer ops) and fix the data path end-to-end:
ownership, definitions, access controls, and freshness.
Then expose that same governed data to AI: search, summarize, and reason over it.
- Productivity gains come from trustworthy context, not better prompts
- Governance solved once at the data layer scales across every AI use case
|
|
|
The Opportunity...
Podcast:
- This week's Open Tech Talks episode 168 is "Building the AI Factory and Lessons on Agentic AI with Maurice McCabe". He has spent 20 years developing systems that ensure things actually work, from scalable SaaS platforms and real-time data pipelines to voice agents deployed in production environments.
Apple | Amazon Music
Courses to attend:
- Document AI: From OCR to Agentic Doc Extraction from DeepLearning. Build agentic document processing pipelines that convert PDFs into structured Markdown and JSON by extracting text, tables, charts, and forms without losing context from layout.
Events:
Tech and Tools...
- OpenCode: The open source AI coding agent.
- Ralph for Claude Code: Ralph is an implementation of Geoffrey Huntley's technique for Claude Code that enables continuous autonomous development cycles
|
|
|
That's it for this week - thanks for reading!
Reply with your thoughts or favorite section.
Found it useful? Share it with a friend or colleague to grow the AI circle.
Until next Saturday,
Kashif
|
|
|
The opinions expressed here are solely my conjecture based on experience, practice, and observation. They do not represent the thoughts, intentions, plans, or strategies of my current or previous employers or their clients/customers. The objective of this newsletter is to share and learn with the community. |
|
|
Dubai, UAE
You are receiving this because you signed up for the AI Tech Circle newsletter or Open Tech Talks. If you'd like to stop receiving all emails, click here. Unsubscribe · Preferences
|
|
|
