Anup Jadhav
Anup Jadhav
Agentic AI Engineering
⌘K
Subscribe Sign in
Anup Jadhav
Anup Jadhav

Anup Jadhav

London
I'm an AI and Software Engineer with 20+ years of technology leadership experience, currently focusing on Generative AI and Agentic AI architectures. I specialize in developing Agentic AI systems that solve complex business problems.

On Durable Objects, Orleans, and prior art for the agentic web

Zak Knill wrote a sharp post this week arguing that LLMs are exposing a gap in our standard cloud-native architecture.

TIL: Ads in AI chatbots are not just a UX problem

TIL from a paper on ads in AI chatbots that putting adverts inside an AI assistant is not the same
Welcome to Middle Loop Engineering
Agentic Architecture

Welcome to Middle Loop Engineering

Where engineering rigour goes now that AI writes the code
How fast does it serve? Throughput, latency, and picking the right GPU
GPUs

How fast does it serve? Throughput, latency, and picking the right GPU

Part 2 of 2 on inference engineering for AI engineers.
Fitting LLMs on Self-Hosted GPUs
LLM Scaling

Fitting LLMs on Self-Hosted GPUs

How much VRAM does your LLM need, and which GPU should you actually rent? A free calculator covering DeepSeek, Llama, Mixtral on H100, B200, A100.
The Harness Is the Product
Claude Code

The Harness Is the Product

Where does product quality live in an LLM-based system? A leaked source and a detailed postmortem, both from Anthropic in the last four weeks, make the answer unusually concrete.

How "Thinking" Models Actually Work

Lilian Weng's Why We Think is a survey of test-time compute and chain-of-thought reasoning. Here's what I pulled out of it.
Harness Engineering: The Outer System That Makes Agents Reliable
Agentic Architecture

Harness Engineering: The Outer System That Makes Agents Reliable

Building a good harness is what separates a good agentic implementation from a great one.

We’re Being Too Loose With the Term “World Model”

I think we are still too loose with the phrase “world model”.

TIL: Quantisation

Quantisation is really a precision-allocation problem.
Write Skills Like Workstations, Not Prompts
Claude Code

Write Skills Like Workstations, Not Prompts

Claude Code skills work best when you treat them as workstations, not prompts: folders with scripts, gotchas, templates, and progressive disclosure that manage the agent's attention budget at runtime.

How the Claude Code team designs agent tools

(Part of my Today I Learned series. Short posts on things that made me think.) When Claude Code shipped, the

Make Claude Code Review Its Own Plans

(Part of my TIL series) If you've used Claude Code's plan mode, you've probably
eval driven development

Ship Prompts Like Software: Regression Testing for LLMs

Because "it seemed fine when I tested it" is not a deployment strategy. Part 4 of 4: Evaluation-Driven Development for LLM Systems
evals

Four Ways to Grade an LLM (Without Going Broke)

Your evaluation technique should match the question you're asking, not your ambition.
eval driven development

Your Golden Dataset Is Worth More Than Your Prompts

Most teams spend weeks perfecting prompts and minutes on evaluation data. That's backwards. Part 2 of 4: Evaluation-Driven Development for LLM Systems
evals

Build LLM Evals You Can Trust

If five correct responses are enough to ship an LLM feature, what are you actually measuring: quality, or luck? Part 1 of 4: Evaluation-Driven Development for LLM Systems

Stripe's coding agents: the walls matter more than the model

(Part of my Today I Learned series) Stripe merges over 1,300 AI-written pull requests every week, and almost every