Kaleb Campbell

Aerospace Engineer, along side Ai building/implementation and engineering.

[Kaleb Campbell @CampbellKaleb23]

Does @thsottiaux hit the reset button anytime today? I just know everyone (Including myself) is dying for a reset. He hasn't reset in over 48 hours I am starting to get worried😂

[Kaleb Campbell @CampbellKaleb23]

Unpopular opinion: one-shotting is not magic. It is specification quality.

Codex can only build the system you clearly define.

The best master prompts combine:

product brief + architecture + test plan + operating contract.

Master prompts becoming the new programming language.

[Kaleb Campbell @CampbellKaleb23]

The Codex vs Claude debate misses the biggest advantage:

A great model with a weak prompt is still guessing.

A master prompt turns intent into an executable contract:

• scope
• architecture
• tests
• constraints
• proof

Any High tier model with a sub par prompt will always under preform than a Low tier model with a master prompt.

Wrong or Right?

[Kaleb Campbell @CampbellKaleb23]

Claude Code can write code.

Codex can run an engineering campaign.

Give Codex a master prompt with architecture, constraints, tests, failure rules and acceptance gates and it stops behaving like autocomplete.

It starts behaving like a team.

What have you experienced by doing this?

[Kaleb Campbell @CampbellKaleb23]

Prompt engineering is being replaced by loop engineering:

Agent attempts.
Live test run.
A verifier rejects weak work.
The agent retries until the result passes.
The model still matters but the harness may matter more.
Is the real AI moat intelligence…or the loop around it?

[Kaleb Campbell @CampbellKaleb23]

What is the point of this now? I have invited 8 friends and they all show accepted in my "Invite a Friend" tab, but I have received 0 resets or anything of the such.

Anyone know what is going on with this and/or if we will ever get a reset for referring a friend?

@OpenAI @OpenAIDevs

[Kaleb Campbell @CampbellKaleb23]

@rezoundous Yes, but I would rather wait an extra hour and use SOL 5.6. Way better confirmation that what you want was built correctly, better debug, better audit, etc. Using master prompts with SOL 5.6 absolutely destroys using master prompts with Fabe 5.

[Kaleb Campbell @CampbellKaleb23]

ChatGPT Work has become the operating layer behind my most ambitious engineering project: building a real-world personal AI assistant that can handle any task under pressure and can research anything that I need it to research then make me a PDF file of it so I can review it.

I use it to turn massive requirements into executable architecture, audit Codex implementations, diagnose hardware and software failures, research technical decisions, and maintain continuity across a project that would normally require an entire engineering team.

What I love most is that it does not stop at giving me an answer it helps transform an idea into a plan, the plan into implementation, and the implementation into something testable.

ChatGPT Work makes one person feel like an engineering organization.

@gdb @OpenAI

[Kaleb Campbell @CampbellKaleb23]

Does @thsottiaux give use a reset tonight?? We shall see.
PLEASE

[Kaleb Campbell @CampbellKaleb23]

@rezoundous Implement arguing when something is wrong or if you have said the wrong thing. Basically like an intent director that does research if you say something that does not make 100% sense.

M A S T E R
P R O M P T S

It will change your life.

[Kaleb Campbell @CampbellKaleb23]

How I use GPT SOL 5.6 to utilize its maximum capabilities:

The best method is to use GPT 5.6 Pro ask it to generate you a master prompt for Codex Sol 5.6, of everything that you want accomplished/built. Make it do (deep research), and at the end of the prompt it also needs to say something along the lines of “when everything is built and implemented run live tests and then if a test fails, go back fix it run another test and then wire everything end to end do not stop until all tests are green.” I have never had an issue with this every time I’ve done this when using Codex Sol 5.6 with master prompts it always double checks works then test everything live and then if anything wasn’t built first on the first try, (which is rarely ever) it will go back and check its work and then retest everything and then it will wire everything end to end inside of the infrastructure of what you are wanting built. Niche tip—> when you are asking GPT SOL 5.6 Pro to create this prompt include: “You are essentially creating the single handedly greatest prompt for the execution style and implementation”. It never fails to amaze me.

What are some tips and tricks you use? I wanna know more on maximizing Sol 5.6

[Kaleb Campbell @CampbellKaleb23]

The best method is to use GPT 5.6 Pro ask it to generate you a master prompt of everything that you want accomplished/built make it do (deep research) and at the end of the prompt it also needs to say something along the lines of “when everything is built and implemented run live tests and then if a test fails, go back fix it run another test and then wire everything end to end do not stop until all tests are green.” I have never had an issue with this every single time I’ve done this when using Codex with master prompts it always double checks works then test everything live and then if anything wasn’t built first on the first try, which is rarely ever it will go back and check its work and then retest everything and then it will wire everything end to end inside of the infrastructure of what you are wanting built. Niche tip—> when you are asking GPT SOL 5.6 Pro to create this prompt include: “You are essentially creating the single handedly greatest prompt for the execution style and implementation”. It never fails to amaze me.

[Kaleb Campbell @CampbellKaleb23]

@rezoundous Sol 5.6 I should clarify.

[Kaleb Campbell @CampbellKaleb23]

Personally Codex 5.6 even on High reasoning absolutely destroys the competition. Using master prompts for each prompt with combining things into one prompt for maximum implementation is the absolute best thing anyone can do when building something. My prediction is that Codex keeps growing and Claude becomes slower in growth because of the Fable 5 restrictions and on 50% of usage. Thoughts? @rezoundous

[Kaleb Campbell @CampbellKaleb23]

@thsottiaux is literally single handedly destroying Anthropic… I have seen so many posts about people canceling there Claude Code subscription and switching to Codex, because they are hitting limits with 1-2 prompts and can't get any work done and only getting to use 50% of Fable 5.
@thsottiaux is basically giving a daily, weekly reset, for every Codex user. When it comes to the the benchmarks benchmarks don’t win users. Products that let people work and build do, @OpenAI @OpenAI @thsottiaux seem to excel particularly in this feild. LITERAL GOATS!

[Kaleb Campbell @CampbellKaleb23]

My prediction:

The Sol 5.6 vs Fable 5 debate will become the new Windows vs Mac.

Sol users will prioritize raw intelligence, tool use, and orchestration.

Fable users will prioritize long-running autonomy and Claude Code workflows.

Which one are you choosing and what convinced you?