A Tryst with Language Models — Subhankar Pattanayak
A Tryst with Language Models
Three years. 110+ editions. 1,700 subscribers. The honest story of how TechSambad was built — one tool, one model, one curious experiment at a time.
by Subhankar Pattanayak · May 29, 2026
Three years ago, I started something I was not entirely sure I could sustain. A newsletter. About AI. Written by someone who, by every honest measure, was a complete beginner. I still am, in many ways. But here we are — 110+ editions later, and this week TechSambad crossed 1,700 subscribers.
I did not plan a milestone post. But sometimes a number stops you in your tracks and makes you look back at the road you have travelled. This is that post. It is not polished strategy. It is just the truth of how this journey unfolded — one tool, one model, one curious experiment at a time.
"The newsletter was never about being an expert. It was about learning out loud — and trusting that someone out there was learning alongside me."
The Journey, Chapter by Chapter
THE BEGINNING · 2023 — Podcasts, Research, and Writing it Down
It started simply. I would listen to AI news podcasts — The Neuron, AI Daily Brief — absorb everything I could, do my own reading, and try to make sense of it all in writing. The newsletter was my way of processing the noise. There was no grand plan. Just curiosity and a blank page.
2023 – 2024 — Discovery: The World of Models Opens Up
Then came yupp.ai — a remarkable platform that gave access to the world's leading models at zero cost. Suddenly I was writing with Claude 3.5, experimenting with GPT-4o, generating images with DALL-E and Black Forest. The newsletter found a new energy. Claude, in particular, became a quiet creative partner — helping me draft, structure, and shape ideas into something readable.
2024 — Manus Enters: Research Gets Agentic
Manus changed something meaningful. For the first time, an agent was doing its own web research — running its own server, pulling its own insights — and handing me a brief I could actually build on. The newsletter stopped being a summary of what I had heard and became a synthesis of what the agents had found. The LLM underneath was still Claude. That felt right.
2024 – 2025 — Google AntiGravity IDE: A Research Partner Arrives
Installing Google AntiGravity IDE was a step change. It became a true research partner — not just a writing assistant, but something that could brainstorm, explore, and help craft articles end to end. Gemini Nano Banana handled the images. The workflow started to feel less like journalism and more like collaboration.
EARLY 2026 — OpenClaw, Hermes, and Building My Own Agents
Early 2026 brought OpenClaw. The installation was not smooth — there were failures, restarts, moments of frustration — but those failures taught me more than any tutorial. With the Hermes platform alongside it, I started building my own agents: ones that would crawl the web, monitor social media, and populate a daily database of AI developments. Codex helped me parse the noise and extract the signal. The AI Brief — my weekly Monday newsletter — was born from this infrastructure. Daily blogs followed on techsambad.com. It had become a small, humming newsroom.
THIS WEEK · MAY 2026 — Claude Code: Full Circle
This week I installed Claude Code. And something felt like coming home. The lessons from OpenClaw — the failures, the mental models built through trial and error — gave me a head start. Within days I had built toolkits for marketing intelligence and bid management. Real tools, doing real work. Claude Code is smart. So is Codex, and AntiGravity, and OpenClaw, and Hermes. Honestly, choosing between them is impossible — each has its strength. What matters is the output: content that is easy to consume, useful to someone, worth the few minutes it takes to read.
What Has Not Changed
Through all of this — the platforms, the models, the agents — one thing has remained constant. I am still learning. Every edition, every experiment, every failed installation is part of that. The newsletter was never meant to be the voice of an expert. It was meant to be the honest account of someone figuring it out in real time, sharing what they found, and hoping it is useful to at least one person out there.
1,700 of you have decided it is. That is not a number I take lightly.
"110 editions. Three years. Every single one written because someone, somewhere, found it worth reading. Thank you for that."
If you have been with me from the early podcast-summary days — thank you for your patience as the format evolved. If you joined recently — welcome to a community that values curiosity over credentials. And if you have never told me what you think of the newsletter — now is a good time. I genuinely want to know what lands, what does not, and what you wish there was more of.
The conversation continues
What do you enjoy about TechSambad? What would you like more of? Reply, comment, or reach out — every response shapes the next edition.
Visit techsambad.com or subscribe on LinkedIn and Substack.
Originally published on LinkedIn · May 29, 2026
TechSambad — curated by Subhankar Pattanayak · AI, thoughtfully curated.