Happy Diwali to you and your family.
I hope the celebrations brought some time with the people who matter to you. Thank you for being part of these conversations, for reading, and for sharing your own experiences with me.
As we get back to work, I want to share something I have been learning by doing.
Over the past few weeks, I have been using AI more directly across research, content, design, prospecting, websites, and product prototypes. The biggest change I am experiencing is how much further I can take an idea myself before handing it over.
That has made me rethink how we organise work, including my own role as a founder.
The distance between an idea and something real is getting shorter
Take the recent work on my website, munesh.coffee.
I could describe the direction, review a layout, ask for changes, improve the writing, and work towards a live result through a conversation with an AI agent.
But the first version was not the finished version. Some layouts needed correction. The mobile view needed attention. Content had to sound like me. Existing articles, URLs, and subscription flows had to be preserved. Cycling statistics needed verification before they appeared on the page.
The work kept moving between building, checking, correcting, and checking again.
That experience matters more to me than a polished demo. It showed me both how much execution has become accessible and how much care a usable result still requires.
When we can produce more, we need to choose better
AI makes it easier to start another project, generate another campaign, or explore another product idea. I can feel that temptation myself.
But the ability to create something does not establish that it deserves our time.
We still have to understand the customer, decide which problem matters, and define what a good result should achieve. A website can look better without making the next step clearer. A proposal can read well without addressing the buyer's concern.
I am finding that a clear brief is becoming a more valuable part of the work: who this is for, what should change, what must be preserved, and how we will know it works.
People can take broader ownership
I do not read this as a reason to value people's contribution less. It makes me more interested in what people can take responsibility for when some of the execution becomes easier.
A marketer may be able to take an idea further into a working campaign. A product colleague may be able to build something customers can react to. A founder may be able to test an assumption before asking a team to invest weeks in it.
Expertise still matters throughout that process. Someone has to recognise a weak assumption, a misleading claim, a security issue, or a customer experience that technically works but feels frustrating.
AI can do more of the work. We still have to decide whether the work is right and take responsibility for what we put into the world.
What the research adds
There is some useful evidence behind this shift. Microsoft's 2026 Work Trend Index surveyed 20,000 workers using AI across ten countries. Among those surveyed, 66% said AI allowed more time for higher-value work, and 86% said they treated its output as a starting point and remained responsible for the thinking.
Those are self-reported findings. They support the direction of my experience, but they do not establish the same result for every team.
That caution matters. In a February 2026 update, research organisation METR explained why selection and measurement problems made its newer developer study unreliable for estimating current productivity gains.
I take that as a reminder to measure the whole job, including review and rework, rather than judge productivity by how quickly the first output appears.
One experiment for the next 30 days
I would start with one recurring piece of work that has a clear outcome. It could be a proposal, a customer update, a research brief, or a small website improvement.
My working approach is simple:
Define. Write down the audience, desired outcome, quality standard, and boundaries.
Build. Use AI for a meaningful part of the workflow and give it enough context to be useful.
Verify. Check facts, usability, and relevant risks. Include correction time in the comparison.
Own. Put a name against the final result, observe what happens, and share what you learn.
At the end of the month, ask whether the result was better, whether it saved effort overall, and whether someone else could repeat the approach.
I am still learning where this works well and where it breaks down. I would particularly like to hear experiences that challenge mine.
My wish after Diwali is that we make room for learning, give people the confidence to try better ways of working, and use any time we gain thoughtfully. There should be room for customers, colleagues, family, and life beyond the next task.
What is one task you can now do with AI that you could not comfortably do a year ago? And what do you still check yourself?
Hit reply and tell me. I would be glad to learn from your experience.
Warm wishes,
Munesh
Cover illustration created with AI. The experiences and judgments in this note are my own.
