Debashish Ghosal

Author profile

Debashish Ghosal

TLDR - Engineer -> Manager -> Tech Leadership, passionate about AI, Developer experience, planet scale infra, transformation - AI-native leadership. Rest here - https://www.linkedin.com/in/deghosal/

@debashish_ghosalSeattle, WAWebsite View on DEV

Articles by @debashish_ghosal

Browse the latest writing surfaced through DevArt.

We Could Have Shipped on Local Models Alone

#ai#llm#healthydebate#testing

We Could Have Shipped on Local Models Alone

Update — v0.1.0 released. CauterRule is now live on GitHub and PyPI. It turns repeated agent...

Debashish Ghosal12 min read

5 reactions0 commentsRead full article

When Your Benchmark Finally Tells the Truth

#ai#llm#agents#healthydebate

When Your Benchmark Finally Tells the Truth

Update — v0.1.0 released. CauterRule is now live on GitHub and PyPI. It turns repeated agent...

Debashish Ghosal8 min read

11 reactions2 commentsRead full article

I Thought Role Separation Would Fix the Optimizer. It Didn't.

#ai#llm#healthydebate#testing

I Thought Role Separation Would Fix the Optimizer. It Didn't.

Previously: 9 Bugs That All Looked Like a Working System · I Built an AI That Rewrites Its Own...

Debashish Ghosal6 min read

11 reactions4 commentsRead full article

I Thought the Optimizer Was the Product. I Was Wrong. The Gate Was.

#ai#llm#healthydebate#testing

I Thought the Optimizer Was the Product. I Was Wrong. The Gate Was.

Previously: 9 Bugs That All Looked Like a Working System · I Built an AI That Rewrites Its Own...

Debashish Ghosal5 min read

9 reactions3 commentsRead full article

I Thought This Was a Classification Problem. It Wasn't.

#ai#healthydebate#llm#testing

I Thought This Was a Classification Problem. It Wasn't.

Previously: 9 Bugs That All Looked Like a Working System · I Built an AI That Rewrites Its Own...

Debashish Ghosal6 min read

11 reactions0 commentsRead full article

I Compared a Local 4B, a Better Cloud Model, and Role Separation. The Results Were Weird.

#ai#healthydebate#llm#testing

I Compared a Local 4B, a Better Cloud Model, and Role Separation. The Results Were Weird.

Previously: 9 Bugs That All Looked Like a Working System · I Built an AI That Rewrites Its Own...

Debashish Ghosal6 min read

12 reactions2 commentsRead full article

My Self-Improving Agent Still Couldn't Improve. That Was the Breakthrough.

#ai#healthydebate#llm#testing

My Self-Improving Agent Still Couldn't Improve. That Was the Breakthrough.

Previously: 9 Bugs That All Looked Like a Working System · I Built an AI That Rewrites Its Own...

Debashish Ghosal7 min read

10 reactions3 commentsRead full article

I Tried 4 Models to Save My Self-Improving Agent. All 4 Failed.

#ai#llm#agents#healthydebate

I Tried 4 Models to Save My Self-Improving Agent. All 4 Failed.

Previously: 9 Bugs That All Looked Like a Working System · I Built an AI That Rewrites Its Own...

Debashish Ghosal7 min read

24 reactions1 commentsRead full article

My Agent Found Real Improvements. The Statistics Still Killed the Promotion.

#ai#agents#llm#healthydebate

My Agent Found Real Improvements. The Statistics Still Killed the Promotion.

Previously: 9 Bugs That All Looked Like a Working System · I Built an AI That Rewrites Its Own...

Debashish Ghosal6 min read

11 reactions2 commentsRead full article

I Let an LLM Rewrite Its Own Prompt. The Real Win Was the Gate That Rejected It.

#ai#healthydebate#llm#agents

I Let an LLM Rewrite Its Own Prompt. The Real Win Was the Gate That Rejected It.

Previously: 9 Bugs That All Looked Like a Working System · I Built an AI That Rewrites Its Own...

Debashish Ghosal6 min read

11 reactions1 commentsRead full article

The Edit That Fixed 4 Tasks and Broke 1

#ai#agents#llm#testing

The Edit That Fixed 4 Tasks and Broke 1

AgentSelfEdit is an open-source sidecar that rewrites its own system prompt from execution feedback....

Debashish Ghosal8 min read

15 reactions0 commentsRead full article

I Built an AI That Rewrites Its Own Prompts — Its Safety Gate Rejected Every Single Edit

#ai#promptengineering#llm#agents

I Built an AI That Rewrites Its Own Prompts — Its Safety Gate Rejected Every Single Edit

AgentSelfEdit is an open-source sidecar that rewrites its own system prompt from execution feedback....

Debashish Ghosal8 min read

24 reactions4 commentsRead full article

9 Bugs That All Looked Like a Working System

#ai#agents#llm#correcting

9 Bugs That All Looked Like a Working System

AgentSelfEdit is an open-source sidecar that rewrites its own system prompt from execution feedback....

Debashish Ghosal10 min read

19 reactions14 commentsRead full article

I Published Every Flaw My Safety Tool Can't Catch. It Made It More Credible, Not Less.

#ai#healthydebate#testing#llm

I Published Every Flaw My Safety Tool Can't Catch. It Made It More Credible, Not Less.

This is a companion to the PlannerCritic series. Article 5 was about what happened when I tried to...

Debashish Ghosal8 min read

10 reactions4 commentsRead full article

My LLM Critic Flip-Flops on Every Run. That's Fine — Because a Frozenset Decides What's Fatal.

#ai#healthydebate#testing#llm

My LLM Critic Flip-Flops on Every Run. That's Fine — Because a Frozenset Decides What's Fatal.

This is a companion to the PlannerCritic series. Article 2 was about a specific critic bug. This one...

Debashish Ghosal7 min read

13 reactions9 commentsRead full article

The Gate That Stayed Silent — When a Blocker Count That Drops Reads as Improvement

#ai#planner#healthydebate#testing

The Gate That Stayed Silent — When a Blocker Count That Drops Reads as Improvement

In the last article, I argued that the safety contract should move out of the LLM critic and into...

Debashish Ghosal5 min read

12 reactions5 commentsRead full article

I Added a Fourth Model Mid-Run. It Changed What My Field Test Could Prove.

#ai#healthydebate#llm#testing

I Added a Fourth Model Mid-Run. It Changed What My Field Test Could Prove.

Latest release: v0.2.2 — Aug 29, 2026 I did something I usually try hard not to do in a field...

Debashish Ghosal7 min read

11 reactions0 commentsRead full article

The Same Model Debating Itself Was More Self-Critical Than Two Different Models

#ai#healthydebate#llm#testing

The Same Model Debating Itself Was More Self-Critical Than Two Different Models

v0.2.1 RELEASED — Aug 28, 2026. Release notes · Field test report v0.2.1 Key Finding: DeepSeek+GPT...

Debashish Ghosal13 min read

14 reactions1 commentsRead full article

Two Projects, One Problem — What PlannerCritic and AdversarialDebate Each Got Wrong

#ai#healthydebate#dogfood#testing

Two Projects, One Problem — What PlannerCritic and AdversarialDebate Each Got Wrong

I built two systems this year that try to solve the same problem from opposite...

Debashish Ghosal10 min read

17 reactions3 commentsRead full article

The Best Model Pair in My Field Test Was Also the Least Trustworthy

#ai#trustworthy#healthydebate#llm

The Best Model Pair in My Field Test Was Also the Least Trustworthy

v0.2.1 RELEASED — Aug 28, 2026. Release notes · Field test report · v0.2.2 release notes v0.2.2...

Debashish Ghosal12 min read

22 reactions7 commentsRead full article

I Thought My Multi-Agent Debate Engine Was Broken. The Real Bug Was the Prompt.

#ai#promptengineering#agents#testing

I Thought My Multi-Agent Debate Engine Was Broken. The Real Bug Was the Prompt.

v0.2.1 RELEASED — Aug 28, 2026. Release notes · Field test report · PyPI v0.2.1 Update: The...

Debashish Ghosal31 min read

15 reactions0 commentsRead full article

Most AI Second Opinions Are Theater. I Built a System That Actually Fights Back.

#ai#healthydebate#testing#llm

Most AI Second Opinions Are Theater. I Built a System That Actually Fights Back.

The first time I ran two LLMs against the same pull request, 89% of their "debate" was fake. Not...

Debashish Ghosal14 min read

15 reactions10 commentsRead full article

My LLM Critic Disagreed With Itself on Every Trial. The Safe Part Was the Code I Didn’t Trust It to Touch.

#ai#healthydebate#testing#llm

My LLM Critic Disagreed With Itself on Every Trial. The Safe Part Was the Code I Didn’t Trust It to Touch.

In the My Agent Refused 96 Times. That Was the Right Output., I argued that the most valuable output...

Debashish Ghosal5 min read

21 reactions8 commentsRead full article

My Agent Refused 96 Times. That Was the Right Output.

#ai#planner#testing#llm

My Agent Refused 96 Times. That Was the Right Output.

In the last article, I wrote about a release story that was weaker than the engine underneath...

Debashish Ghosal10 min read

24 reactions5 commentsRead full article

Most AI Second Opinions Are Fake. I Built a Two-LLM Review Engine to Prove It.

#ai#adversarial#llm#testing

Most AI Second Opinions Are Fake. I Built a Two-LLM Review Engine to Prove It.

Most AI "second opinions" are fake. Not because there is no second model. Because the second model...

Debashish Ghosal11 min read

17 reactions12 commentsRead full article

A Reader Audited My OSS Release in Public. He Found the Contradictions I Missed.

#ai#planner#llm#testing

A Reader Audited My OSS Release in Public. He Found the Contradictions I Missed.

When I shipped v0.2.1 of PlannerCritic, I thought the hard part was over. The engine had survived a...

Debashish Ghosal7 min read

18 reactions8 commentsRead full article

I Tried to Prompt-Inject My Own Agent Engine. It Didn't Work. Here's Why.

#ai#agents#security#llm

I Tried to Prompt-Inject My Own Agent Engine. It Didn't Work. Here's Why.

This is article 5 in a series about building PlannerCritic, an open-source engine where one LLM...

Debashish Ghosal12 min read

41 reactions11 commentsRead full article

I Ran 170 Agent Goals for $0.49. The Field Test Found 10 Issues That Unit Tests Never Would.

#ai#llm#infrastructure#testing

I Ran 170 Agent Goals for $0.49. The Field Test Found 10 Issues That Unit Tests Never Would.

This is article 4 in a series about building PlannerCritic, an open-source engine where one LLM...

Debashish Ghosal13 min read

17 reactions5 commentsRead full article

The Planner Made the Same 3 Mistakes Every Time. A Bigger Model Didn't Fix It.

#ai#agents#llm#testing

The Planner Made the Same 3 Mistakes Every Time. A Bigger Model Didn't Fix It.

This is article 3 in a series about building PlannerCritic, an open-source engine where one LLM...

Debashish Ghosal5 min read

21 reactions28 commentsRead full article

I Told My LLM Critic to Be Adversarial. It Started Blocking Plans for Being 'Not Thorough Enough.'

#ai#adversarial#llm#promptengineering

I Told My LLM Critic to Be Adversarial. It Started Blocking Plans for Being 'Not Thorough Enough.'

This is article 2 in a series about building PlannerCritic, an open-source engine where one LLM...

Debashish Ghosal4 min read

16 reactions26 commentsRead full article

Debashish Ghosal developer articles | DevArt