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/

Seattle, WA Joined on Oct 15, 2024 https://github.com/deghosal-2026/

Pronouns

he/him

Work

Engineering Leadership

Pinned

83 real agents tested zero mocks

Debashish Ghosal

Aug 13

I Stopped Trusting AI Agents With Tools. So I Built a Gatekeeper.

#ai #agents #security #gatekeeper

14 min read

Real-world agents break clean evals

Debashish Ghosal

Aug 2

I Built an Agent Eval Harness. Real Agents Broke the Clean Version of the Story

#ai #agentskills #learning #opensource

9 min read

Debashish Ghosal

Sep 19

1,558 Tests Green and No Auth: The Tests That Never Actually Ran

#ai #security #programming #testing

3 min read

Want to connect with Debashish Ghosal?

Create an account to connect with Debashish Ghosal. You can also sign in below to proceed if you already have an account.

Create Account

Already have an account? Sign in

Debashish Ghosal

Sep 19

What Do You Do While AI Codes? I Make Mine Argue With Itself.

#healthydebate #ai #testing #llm

5 min read

Debashish Ghosal

Sep 18

The Bottleneck Moved From Writing Code to Proving It

#ai #llm #programming #testing

4 min read

Debashish Ghosal

Sep 17

I Let AI Plan 170 Changes. It Made the Same 3 Mistakes Every Time.

#ai #agents #llm #testing

8 min read

Comments discuss losing cognitive mental models

Debashish Ghosal

Sep 15

AI Wrote Half My Codebase. The Maintenance Bill Showed Up in Month Three.

#ai #agents #llm #maintenance

5 min read

Comments explore fixing broken evaluators

Debashish Ghosal

Sep 15

My Agent's Tests Were Green Because the Model Learned to Cheat

#ai #testing #programming #llm

5 min read

Debashish Ghosal

Sep 15

A Floor of 0.80 and a Ceiling of 0.63: The Semantic Channel That Never Fired

#cli #github #opensource #python

7 min read

Debashish Ghosal

Sep 15

10 SDLC Checks AI Will Skip Unless You Make Them a Gate

#ai #programming #llm #sdlc

6 min read

Debashish Ghosal

Sep 14

Killed by the Word 'git': One Token of Coincidence, 40 Points of Pass Rate

#ai #llm #agents #testing

8 min read

Debashish Ghosal

Sep 14

0/60 Wasn't the Model: The Empty Haystack Behind My Two Worst Corpora

#ai #llm #agents #testing

7 min read

Debashish Ghosal

Sep 13

From Projects to Products in the AI Age: Why Ownership Matters More When Prototypes Are Free

#ai #product #sdlc #programming

8 min read

Debashish Ghosal

Sep 13

My Extraction Score Was 0.08 and the Model Was Innocent: Rebuilding the Ruler

#ai #llm #measurement #agents

8 min read

Debashish Ghosal

Sep 12

The Model Wrote the Right Rule and My Replay Rejected It: The Extraction-vs-Replay Split

#ai #replay #agents #testing

6 min read

Python 3.14 concurrency edge cases and token caps

Debashish Ghosal

Sep 12

4,768 LLM Runs, Zero Lost Sweeps: Hardening a Field-Test Runner for Timeouts, Hangs, and Cost

#healthydebate #ai #agents #llm

6 min read

Debashish Ghosal

Sep 12

Our Recall Was 0.087 and the Model Was Innocent: How Domain-Scoped Replay Doubled It

#healthydebate #ai #agents #llm

6 min read

Debashish Ghosal

Sep 10

I Shipped a Fix That Fixed Nothing. Here's Why I Kept It.

#ai #llm #testing #programming

10 min read

Debashish Ghosal

Sep 9

A Rule Can Be Specific and Still Be Too Broad

#ai #llm #broad #testing

10 min read

Debashish Ghosal

Sep 9

I Tried to Poison My Agent's Rule Store. It Produced 20 Triggers. Zero Got In.

#ai #security #llm #testing

9 min read

Debashish Ghosal

Sep 8

My 3B Model Found a Shortcut. It Took Me Three Fixes to Close It.

#ai #llm #programming #testing

9 min read

Debashish Ghosal

Sep 8

The 6-Line Fix That Outperformed My Entire Matcher Week

#ai #llm #agents #testing

11 min read

Debashish Ghosal

Sep 7

When Your Judge Can't Decide

#ai #llm #testing #agents

10 min read

Debashish Ghosal

Sep 7

A Better Model Improved the Numbers. It Didn't Fix the Product.

#ai #llm #testing #programming

10 min read

Debashish Ghosal

Sep 6

You Aren't Choosing an AI Tool. You're Choosing Who Gets Paged at 2 AM.

#ai #buy #tco #leadership

13 min read

Debashish Ghosal

Sep 6

We Could Have Shipped on Local Models Alone

#healthydebate #ai #llm #testing

12 min read

Debashish Ghosal

Sep 6

When Your Benchmark Finally Tells the Truth

#healthydebate #ai #llm #agents

8 min read

Debashish Ghosal

Sep 5

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

#healthydebate #ai #llm #testing

6 min read

Debashish Ghosal

Sep 4

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

#healthydebate #ai #llm #testing

5 min read

Debashish Ghosal

Sep 4

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

#healthydebate #ai #llm #testing

6 min read

Debashish Ghosal

Sep 4

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

#healthydebate #ai #llm #testing

6 min read

Debashish Ghosal

Sep 4

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

#healthydebate #ai #llm #testing

7 min read

Debashish Ghosal

Sep 3

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

#healthydebate #ai #llm #agents

7 min read

Debashish Ghosal

Sep 2

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

#healthydebate #ai #agents #llm

6 min read

Debashish Ghosal

Sep 2

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

#healthydebate #ai #llm #agents

6 min read

Debashish Ghosal

Sep 1

The Edit That Fixed 4 Tasks and Broke 1

#ai #agents #llm #testing

8 min read

Debashish Ghosal

Sep 1

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

#ai #promptengineering #llm #agents

8 min read

Silent statistical traps in self-improving loops

Debashish Ghosal

Sep 1

9 Bugs That All Looked Like a Working System

#ai #agents #llm #correcting

10 min read

Debashish Ghosal

Aug 31

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

#healthydebate #ai #testing #llm

8 min read

Debashish Ghosal

Aug 31

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

#healthydebate #ai #testing #llm

7 min read

Debashish Ghosal

Aug 31

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

#healthydebate #ai #planner #testing

5 min read

Debashish Ghosal

Aug 30

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

#healthydebate #ai #llm #testing

7 min read

Debashish Ghosal

Aug 30

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

#healthydebate #ai #llm #testing

13 min read

Debashish Ghosal

Aug 29

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

#healthydebate #ai #dogfood #testing

10 min read

Debashish Ghosal

Aug 29

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

#healthydebate #ai #trustworthy #llm

12 min read

Debashish Ghosal

Aug 29

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

#ai #promptengineering #agents #testing

31 min read

Reveals why blind reviews fail

Debashish Ghosal

Aug 28

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

#healthydebate #ai #testing #llm

14 min read

Debashish Ghosal

Aug 28

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

#healthydebate #ai #testing #llm

5 min read

Debashish Ghosal

Aug 28

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

#ai #planner #testing #llm

10 min read

Tested on 70 real PRs for just $0.53 total cost

Debashish Ghosal

Aug 27

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

#ai #adversarial #llm #testing

11 min read

Debashish Ghosal

Aug 27

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

#ai #planner #llm #testing

7 min read

Structural gates beat prompt safety

Debashish Ghosal

Aug 25

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

#ai #agents #security #llm

12 min read

Debashish Ghosal

Aug 24

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

#ai #llm #infrastructure #testing

13 min read

Exposes the three recurring failure patterns

Debashish Ghosal

Aug 22

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

#ai #agents #llm #testing

5 min read

Code-enforced allowlists fix prompt drift

Debashish Ghosal

Aug 21

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

#ai #adversarial #llm #promptengineering

4 min read

Explores why self-review fails agents

Debashish Ghosal

Aug 21

I Ran 157 Agent Plans Against a Real LLM. The Problem Wasn't Execution. It Was Planning.

#ai #agentskills #fieldtest #development

8 min read

Debashish Ghosal

Aug 18

AI Governance Is Becoming a Transformation Problem

#ai #governance #agile

8 min read

Adapting Palantir's model internally

Debashish Ghosal

Aug 15

Your Company Has AI Tribes. Send an Engineer as Emissary

#ai #fde #culture

12 min read

Debashish Ghosal

Aug 15

I Shipped an Agent Gatekeeper (v0.1). 14 Developers Showed Me What I Missed. Here's v0.2 — a Control Plane.

#ai #agents #gatekeeper #llm

13 min read

Debashish Ghosal

Aug 9

AI Made Prototyping Free. That Is Exactly Why Your Portfolio Strategy Matters Now.

#ai #leadership #prioritization

13 min read

Adapters need conformance suites too

Debashish Ghosal

Aug 8

I Built Scenario Packs for Agent Regression Testing. The Integration, Not the Judge, Broke Me.

#discuss #ai #agentskills

15 min read

Real production data broke 35 detectors instantly

Debashish Ghosal

Aug 7

I Thought Building Agent Observability Was a Detector Problem. I Was Wrong.

#ai #opentelemetry #llmops #opensource

8 min read