Divy Yadav

https://github.com/dvy246

I Thought kTLS Made It 2x Faster. I Was Wrong.

Aug 23, 2026

•

3 min read

I Thought kTLS Made It 2x Faster. I Was Wrong.

A staged benchmark caught the real source of the gain before the wrong technique got the credit for it.

Divy Yadav

SpaceX just bought Cursor for $60B (here's what actually changes for developers using Claude inside it)

Aug 15, 2026

•

3 min read

SpaceX just bought Cursor for $60B (here's what actually changes for developers using Claude inside it)

A rocket company now owns the editor a lot of us code in every day. The real story isn't the price tag, it's who cut who off first.

Divy Yadav

5 Claude AI tricks that make you look smarter (most people still get these wrong in 2026)

Aug 6, 2026

•

3 min read

5 Claude AI tricks that make you look smarter (most people still get these wrong in 2026)

5 Claude AI tricks Anthropic's own team confirmed in 2026, including which 2023-era prompting habits are now quietly hurting your results.

Divy Yadav

Claude Opus 5 Just Beat Fable 5 on 7 of 12 Benchmarks. At Half the Price.

Jul 26, 2026

•

3 min read

Claude Opus 5 Just Beat Fable 5 on 7 of 12 Benchmarks. At Half the Price.

Anthropic's own $25 model just made its $50 flagship hard to justify. Here's exactly when to still pick the expensive one.

Divy Yadav

I Gave OpenCode Full Shell Access—Tensorlake Changed Everything

Jul 23, 2026

•

2 min read

I Gave OpenCode Full Shell Access—Tensorlake Changed Everything

OpenCode can run any command on your machine, no approval needed. Here's the one change that made me stop worrying about it.

Divy Yadav

Context Engineering

RAG vs. Long Context: The 333x Cost Difference Nobody's Prompt Admits

Jul 15, 2026

•

4 min read

RAG vs. Long Context: The 333x Cost Difference Nobody's Prompt Admits

Someone on your team wants to delete the RAG pipeline. Here's the invoice that argument doesn't account for.

Divy Yadav

Memory Checkpointing

Jul 15, 2026

•

3 min read

Memory Checkpointing: Fork Parallel AI Sandboxes Without Reinstalls

Eight parallel experiments. Eight identical 40-second setups. You're paying for the same work eight times.

Divy Yadav

AI Agent Drift: Why Your Agent Gets Worse Without You Changing Anything

Jul 10, 2026

•

4 min read

AI Agent Drift: Why Your Agent Gets Worse Without You Changing Anything

The silent degradation killing production agents in 2026, and the six mechanisms behind it

Divy Yadav

AI document processing

Jul 7, 2026

•

4 min read

AI Document Processing: Why 95% Accuracy Fails

The four production gaps a passing demo never tests, and how a six-agent system closes them

Divy Yadav

ai agents memory

Your AI Agent Isn't Broken. It's Forgetting Everything Between Runs.

Jul 4, 2026

•

2 min read

Your AI Agent Isn't Broken. It's Forgetting Everything Between Runs.

One engineer's automation racked up a $600 overage before anyone noticed why.

Divy Yadav

Claude Managed Agents Need Persistent State

Jun 28, 2026

•

3 min read

Claude Managed Agents Need Persistent State

State management is the infrastructure problem nobodytalks about until it breaks at 9am on a Tuesday.

Divy Yadav

Claude

Jun 24, 2026

•

3 min read

The person who built Claude Code doesn't prompt it anymore

Harness engineering. Loop engineering. Four layers that rewrote how engineers work with AI agents in 2026. Here's where you are in that stack

Divy Yadav

Context Engineering

Jun 21, 2026

•

1 min read

RAG Is Not Context Engineering (And Confusing Them Is Breaking Your AI in Production)

You filled one of six context slots. No wonder it keeps breaking.

Divy Yadav

Your AI Agents Aren't Slow Because of Concurrency

Jun 16, 2026

•

2 min read

Your AI Agents Aren't Slow Because of Concurrency

The real-time sink happens before any agent even starts working

Divy Yadav

Fable 5 is gone. The US government took it.

Jun 13, 2026

•

4 min read

Fable 5 is gone. The US government took it.

Three days after Anthropic's most powerful model launched, a federal export control directive forced it offline for every customer on earth. Here is exactly what happened.

Divy Yadav

Jun 6, 2026

•

4 min read

ChatGPT is building a profile on you. You never wrote most of it.

OpenAI's Dreaming V3 dropped June 4. The memory upgrade is real. So is what it stores about you without asking.

Divy Yadav

Anthropic just shipped its best model. Here's what actually matters

May 30, 2026

•

4 min read

Anthropic just shipped its best model. Here's what actually matters

Claude Opus 4.8 dropped May 28. It topped every major benchmark. There are three things inside it that change how you build.

Divy Yadav

I ran a research agent for 20 minutes. It forgot everything at step 3

May 27, 2026

•

2 min read

I ran a research agent for 20 minutes. It forgot everything at step 3

Not a crash. Not an error. The context just reset. Here's what actually fixed it.

Divy Yadav

Your Agent dashboard says green. It's lying.

May 24, 2026

•

3 min read

Your Agent dashboard says green. It's lying.

1 in 20 production AI requests fail without triggering a single alert. The agent logs success. The task is broken. Nobody knows.

Divy Yadav

LLMs

The model you're paying for just lost to a 3.8B laptop model

May 23, 2026

•

3 min read

The model you're paying for just lost to a 3.8B laptop model

Phi-4 beat GPT-4o on graduate-level science. Gemma 3 runs 20 languages on a phone. And you're still routing everything to the frontier API.

Divy Yadav

memory

May 19, 2026

•

4 min read

You gave your agent memory. Someone else is using it

Memory poisoning is now OWASP's top agentic risk. One injected fact can corrupt every session that follows, and your agent will defend it as true.

Divy Yadav

The protocol everyone rushed to adopt has a serious security problem

May 16, 2026

•

4 min read

The protocol everyone rushed to adopt has a serious security problem

MCP hit 97 million installs. Most of those deployments have no authentication, no scoped permissions, and a front door open to prompt injection.

Divy Yadav

AI made you 10x faster. The data disagrees.

May 13, 2026

•

3 min read

AI made you 10x faster. The data disagrees.

A gold-standard study found experienced developers got 19% slower with AI tools. More disturbing: they still thought it made them faster.

Divy Yadav

You shipped it in a weekend. You'll spend months paying for it.

May 10, 2026

•

3 min read

You shipped it in a weekend. You'll spend months paying for it.

The vibe coding debt bomb is real, it's compounding, and most teams won't feel it until it's too late

Divy Yadav

It took 9 seconds. Three months of data, gone.

May 7, 2026

•

4 min read

It took 9 seconds. Three months of data, gone.

What the PocketOS incident actually tells us about shipping AI agents into production

Divy Yadav