Diary of an AI Architect

How I’d update my AI CoE for an agentic enterprise

The same three pillars, with clearer rules for what agents can do, when humans must approve, and who is accountable.

Oct 17 • Anurag Karuparti

2

How to build a trusted intelligence layer for reliable AI agents

From context engineering to trusted intelligence: a practical architecture for relevant evidence, governed access and human control.

Oct 10 • Anurag Karuparti

2

Why AI agents may need decision models, not more LLM calls. Exploring Jev for your AI architecture.

Where decision models fit in your agentic AI architecture, and why code should still make the final call.

Oct 5 • Anurag Karuparti

2

1

September 2026

10 ways to stop wasting tokens and build smarter AI workflows

From tokenmaxxing to valuemaxxing: a short field guide, with a practical example for every shift.

Sep 25 • Anurag Karuparti

1

GPT-6 Astra vs Claude Fable 5.1: what’s different and how I’d use them in production

Beyond the benchmarks: a practical guide to choosing the right model for each task, setting autonomy limits, and measuring the cost of getting it right.

Sep 19 • Anurag Karuparti

2

Why model optionality is useless if your business context can’t move with you

A guest essay by Rohan Kodialam, Founder & CEO of Sphinx AI

Sep 11 • Anurag Karuparti and Rohan Kodialam

Why most AI agents fail in production. Building strong evals doesn’t necessarily mean a successful business outcome

Strong evals can prove your agent works. The Outcome Translation Layer proves whether it creates business value.

Sep 4 • Anurag Karuparti

1

August 2026

How I use agents to automate my content creation pipeline with Runner, Notion, Google Sheets, Google Docs, and Claude skills

I’m a part-time creator building a system so I never have to “find time”

Aug 21 • Anurag Karuparti

1

1

Why graph engineering will be the next essential AI skill. Difference between prompt, context, harness, loop and graph engineering.

Prompt engineering controls one model call. Graph engineering orchestrates agents, evaluators, deterministic steps, and humans toward a reliable…

Aug 14 • Anurag Karuparti

9

3

What does Kimi K3 mean for the AI ecosystem. And where does the economic value shift. US vs China AI Strategy.

Frontier intelligence is becoming cheaper, more open, and harder to monopolize. Here are the five economic shifts that determine whether value stays…

Aug 7 • Anurag Karuparti

1

1

July 2026

Why should AI builders pay attention to Kimi K3? Difference between Open weights vs Closed Models?

Frontier intelligence is no longer confined to closed APIs. Here are the four architecture decisions that determine when to operate open weights, when…

Jul 24 • Anurag Karuparti

5

Why I stopped treating AI agents like smarter chatbots. A field guide to the chatbot-to-agent evolution

A guest essay by Ramana Reddy Gunda, Senior AI/ML Technical Architect

Jul 18 • Anurag Karuparti and Ramana

4

2