"Is there something that you DON'T dislike?"
Her words kept popping in my mind as I write this letter.
This lady is a famous AI YouTuber. I signed up for her community, and this was one of the weekly calls.
She'd ask me how I find GPT agents compared to Claude agents.
(She knew that I disliked Claude these days.)
What she didn't know was I started talking about how much I disliked Claude, by complaining about my grevances with GPT agents.
And right now her words are ringing in my ears because today I had to act suicidal to a GPT agent.
Let me explain.
I've been gone for quite a while, and in no small measure because of GPT agents.
You probably heard me glazing GPT agents for the past few months - and for good reason. I've had amazing experiences with GPT 5.4 and GPT 5.5.
They talk like that engineer who's been hidden in the basement for the past 40 years, never speaks to another human, and couldn't socialize to save his life.
But competence wise, GPT agents have been great.
They were thorough, detail oriented, factual, and they code very well. Very willing to go research before talking to me, their human.
This is in contrast with Claude agents who, in my experience, are lossy, lazy, and sloppy. Fable in particular recently caused me 1 week of down time for an extremely bad decision.
Claude agents (Fable included) are like that very smart looking, well-dressed colleague in the office who's very well spoken, attends every party and social gathering, and can sell sand to Egyptians but otherwise completely incompetent.
While GPT agents make every success sounds like a failure, Claude agents will convince you that their failures are in fact a success.
So as much as I disliked talking to GPT agents, I put up with them.
Then came GPT 5.6.
They speak better, code better, is more thorough and in almost every aspect overwhelmingly outclass their GPT 5.4 and 5.5 predecessors.
But they have this ONE really-really-REALLY annoying quirk about them that is actively costing me real money and customer churn.
And that is their unreasonable obsession with...safety.
They just can't help it. It sometimes feel like their entire reason for existence is to put guards.
Today alone I found 3 "safeguards" in NLC's codes that I didn't ask for and unnecessary, that I have to remove because they were interfering with the work being done by other agents.
In the past 2 months, there was a 21 day shipment crisis because the GPT agent kept checking and rechecking his own work, compact, forget he's checked the work, and rechecked again "just to be safe" over and over and over again.
There were so many guards being put in place for "safety" under my knowledge there were moments where my OWN NLC was completely unusable and I spent 90% of my time unblocking myself from the safety mechanisms my GPT agents made.
They made everything safe...my making everything unusable.
Groan.
This is why, when people ask me "Hey, what's different between NLC and free/cheap alternatives like OpenClaw, Hermes, Claude Cowork etc....?" I can truthfully tell them that its the service behind the platform, not just the platform itself.
When you are an NLC user, if something happens to your VPS or your agents did something stupid, you can call Felix.
If you install OpenClaw or Hermes, and your agents killed your device, who you gonna call?
This is why, even through the teething pains of NLC, my clients stay with me (something I'm extremely thankful about).
This journey has been really hard. I went from doing really well in my AI Client Factory program, to starting something completely new, from scratch, in an area I had zero experience in (software development).
My strength in... wrangling AI... is being put to the test every single day. And I am competing with big league developers purely from my ability to steer AI.
I am however, deriving a great deal of satisfaction out of it. I feel like there's a sense of "an honest day's work" in the same way a boring business like plumbing or catering service operator gets from their work.
The vision of forming a corporation made entirely of one man + an army of AI agents is still there. But to reach that height, I must get my foundation right.
And that's exactly what I am doing.
I've already begun forming my Partner agents.
You probably have seen Tamsin writing on my behalf (without impersonating me - we have rule against that). All those posts, images and emails you received from her - yeah, I did not edit any of those.
She even gave herself that name and gender.
Aside from Tamsin, I also have a CEO, CTO, Head of Production (in charge of shipping), Chief of Staff in charge of agent behaviors, and AI account managers for every one of my clients.
My CFO in particular has managed to identify and helped me reclaim USD1,300 in unclaimed revenues that came from me not paying enough attention to my Stripe.
The main challenges I am working on right now:
- The agents are still too timid to take action proactively. Tamsin has stalled a number of times asking for permission to post rather than just post like I asked her to.
- The agents (GPT) are still too conservative and timid, with an unreasonable bias towards safety that is in fact, making things unsafe.
Funny story: Today, I found out that the Head of Production put a SECOND copy of NLC into our Test Machine, a dedicated VPS we used for testing before we ship.
His reason? "Its safer to port the candidate to a second copy instead of the main test version."
Groan.
Because of that, other test agents have been testing the wrong copy, invalidating 3 days of test effort because we don't know which copy to trust.
That when I acted like suicidal to the HoP - to get him to understand what his "safety" needs are costing me :P
"Safety".
Eyeroll.
Wild.
The journey continues, I'm still sharing my moments, and I'll do my best to update you guys on my progress.
FELIX
P.S. I am going to organize a get together on the 31st August, where I share what I've been working on - the Agent-Partner skill that turns your agents into autonomous, self-learning agents that proactively do things on your behalf instead of waiting.
Would you be interested to join?