Daniel Lynch

Chasing my dreams one line of code at a time.

[Nick shirley @nickshirleyy]

CBS News said there was no evidence of fraud.

The NYT said the Somali community was being targeted

CNN said there was "little evidence."

Tim Walz said it was “white supremacy” to expose fraud

Today: $90M busted and 15 charged.

IT WAS ALL FRAUD AND THEY KNEW.

[Daniel Lynch @Ambitious_Lynch]

Dynamic tool provisioning using mesh of pnpm vs npm vs go pip pipx pypi and registries fine tuning to stop reinventing the wheel and become to cli operations like a human would to use the guardrails of the cli and terminal shell and have cataloging for all cli installations available vs agent slosh of packages and resources already available open source and understanding to define surfaces, planes, layers using finite element modeling to parse and use transformations of all json toml yaml toon files with jq yq taplo to use jacobian gaussian elimination and n=5 p<0.1 tail end binomial distribution with markov chains, partial diff eq etc and create systems of equations for unique solutions. I have been trying with java x rust x python but the drift is so hard to contain without overconstraining the system while maintain proper degrees of freedom

[Daniel Lynch @Ambitious_Lynch]

Boris first off love claude code absolutely will never stop using it, but gpt 5.5 vs opus 4.7, honestly gpt 5.5 crushed the game. I had cancelled my openai subscription and with the codex app and 5.5 i really love how well it understands the intent, the nlp is dead on and it seems to understand the “goal” so much more dead on that opus does, and even without skills or mcp, raw dawging with gpt 5.5 you go so much further with your token and the ability to intuitively understand when to properly call tools. Using gpt 5.5 with pi coding agent as the harness is honestly unbelievable what u can do with little effort, login and sync subscription, it autodetects all skills and I have been using all models but gpt 5.5 as the driver until you guys clawd your way back on top with next model i am sure you already have lined up.

[Daniel Lynch @Ambitious_Lynch]

6 months abroad and funny how everyone I speak at home thinks Im crazy for living in airbnbs in new places I have never been.

Yet I have NEVER in my life:
- Been happier
- Made so much $$$
- Found true purpose
- Realized how closed minded I was before

Invited by an oil & gas CEO for a new project for biggest deal of my life, booking private plane and wants me to help treat 500 cancer patients on their dime.

Started a new business, did 6 figures in first 30 days, and landed an investor fo a second location, with no active marketing even though I have owned an agency for the last decade.

So I can’t help but laugh when people who have never gone out their comfort zone try to convince and then say Not for me, I would never, too dangerous, etc.

And in my head on repeat I think - there is a reason why you would rather tell me what doesnt work for you vs realize you dont have the opportunity because you would rather live in fear vs see else is out there in the world.

#mexico

[Magee Clegg @mageeclegg]

One of the strangest things about America 🇺🇸…

You meet people making $300k/year who are anxious, exhausted, medicated, and can barely sleep at night.

Then you go to Mexico 🇲🇽…

and see some guy running a small restaurant, barely breaking even…

but he’s laughing with friends, drinking mezcal at lunch, hugging customers, and sleeps perfectly fine at night.

[Daniel Lynch @Ambitious_Lynch]

@_javierivero @cursor_ai Representing Mexico City - where do we start building magic together

[Daniel Lynch @Ambitious_Lynch]

@RighttoTryGuy Natural killer cells and cytotran in Mexico City

[Daniel Lynch @Ambitious_Lynch]

@bullishbruk @SecKennedy @realDonaldTrump @DrMakaryFDA It is available at Pathways to Heal in Mexico City

[Daniel Lynch @Ambitious_Lynch]

@thsottiaux Numerical methods and finite element modeling to define systems with degrees of freedom, constraints, and jacobian gaussian elimination using toml, yaml, json file to define systems of matrics that can define various key values, parameters, etc.

[Daniel Lynch @Ambitious_Lynch]

@DrMakaryFDA @DrPatrick Been doing NK in Mexico for my lyme disease, has worked wonders vs the USA I was a pin needle for IVS.

https://t.co/42Edn0gf3d

[Daniel Lynch @Ambitious_Lynch]

@DawnsMission I am doing NK in Mexico for my Lyme https://t.co/42Edn0gf3d

[Daniel Lynch @Ambitious_Lynch]

Check out Mexico for Nk treatment plenth of clinics, have worked in healthcare and holistic/integrative/alternative the next barrier after FDA will be insurance coverage, access due to supply and demand, while have worked with over a hundred patients who did immunotherapy within days.

[Daniel Lynch @Ambitious_Lynch]

Every American man should be a passport bro before marriage just to see how woman in other countries treat men.

You will never believe a woman who speaks multiple languages, has more education, and is actually independent vs claims to be could ever compare until you see the light.

#PassportBros #mexico #latina

[Daniel Lynch @Ambitious_Lynch]

Mad respect great pivot, not doubt you shall rewarded by OSS community. I feel the key to SaaS now to prevent being cannabilized by the growing reality that AI will only move quicker is target specialized niches with turn key solutions I have been focus on cash pay holistic and integrative medical providers for a decade and now anthropic aligned w me on that wss their push on life sciences

[Daniel Lynch @Ambitious_Lynch]

I feel a large aspect of peformance degradation has to do with codebase blost and technical debt, that cylically after certain thresholds of developer you need to programtically clear all caches for all package managers, build artifacts, node modeuled and audit for stale lock files, and then in addition the claude cache, logs, verbose plus anything else
Keeps chipping away at the amount of resources claude needs to keep track of unless ignored but even then I have found with claude plugins they also have caches. I have if i Claude is on struggle bus and i uninstall reinstall refresh all of above, it resolves a lot of performance degradation, just super painful to do it and in midst of the grind.

Cursor is even worse with the vacuum hugiene needs on all the sql datasbses that quickly explored to 4+ gb and it makes a backup so all of a sudden u have 8 gb plus logs bloat and verbose bloat.

[Daniel Lynch @Ambitious_Lynch]

Love both you guys, @dharmesh a few years ago at inbound you said to not chase ai you will never win but dig a moat, so i got my shovel and started mastering the data ecosystem and i would say when opus 4.1 hit i got to finally start enjoying my data moat which they coined a “data lake” and replace my entire staff w ai.

[Daniel Lynch @Ambitious_Lynch]

@alex_prompter Claude

[Daniel Lynch @Ambitious_Lynch]

@tom_doerr Legit game changer almost like they should just eliminate codex and only use gpt 5. Also discovered if you evaluate your brewfile and clitools, tell it to only use posix, modern cli tools, vendor cli tools, and do gap analysis for each task, it does well.

[Daniel Lynch @Ambitious_Lynch]

I was historically team claude code with backup as amazon q. I have exponentially increased using codex, your team is doing great job!

I have found codex excels at hallucinating less and is better at tool calling out of the box and evaluating more technical deep dived that are focused specific objectives.

The o3 models is so solid at planning and better than claude code, wouod
Be awesome to have that mode available and deep research bundled together to provide better self orcestration.

Claude codex excels w this that when i am using parallel agents with neural networks and defined objectives i can have autonomously self orchestrate with judge llm / MoE and integrated with ollama
Local llm models i use oss-120b via turbo mode for cloud inference and then spawn 100 agents with DAA and configure TDD and HDD (hypothesis driven development) and require scinetific method with judge llm and let it rip for 3 hours autonomously w high levels of accuracy and limited hallucinations using a recurisve closed feed back loops based on defining a mathemtical function for each objective to calculate area under curve using riemmans sums and tell it to approsch infinity defining infestimally small increments based in newton raphson approximation.

[Daniel Lynch @Ambitious_Lynch]

Prob same people who blame themselves for being fat eating donuts all day. Claude code shits on codex all day to infinity and beyond, it is not even a comparison. Anyone who disagrees prob also voted for joe biden and thinks you can build rome in a day one shot prompting and forgetting to do all the environmental setup and configurations.

No matter what coding agent, everything is about setting up the proper shell environment, tools and tool call infrastructure, and using mcp serverd, and what i have done w claude code blew my mind.

Codex has its place and before gpt 5 didnt even bother w it, but still light years behind.

Codex in Cursor IDE though is a lot better, almost hard to think it is the same tool and does a lot better w tool calling.