Logan

Dad. Husband. Cybersecurity Engineer. Co-founder @ Building managed AI solutions for SMBs @

[Logan @itsLoganAbell]

Does codex feel faster than usually today?

[DHH @dhh]

For anyone with endless ideas, this agent age is nirvana as those ideas are met with endless execution, endless exploration. I've never had has much fun working with computers as I do right now. What a time to be alive.

[Logan @itsLoganAbell]

@IQReactorAI @karibriski How much computer you are going to get out of that?

[Logan @itsLoganAbell]

@yacineMTB And going to bed thinking about what prompts you’ll start off with in the morning.

[Logan @itsLoganAbell]

@Da7_Tech Tough seeing the grok usage in Hermes is bad. I was hoping to try it out since it’s supposed to be a fast model that benchmarks well. For general purpose and orchestration, if you were to pick one model, what would it be?

[Logan @itsLoganAbell]

@quant_xbt Same here, if you’re a heavy user the limits on Cursor with grok 4.6 high fast are disappointing

[Logan @itsLoganAbell]

@mrsharma Hermes also has a fantastic team behind them, constantly pushing out new features and improvements. Hermes can do everything these Mastra wrappers can do plus more.

[Logan @itsLoganAbell]

Unlike previous gen models in the gpt family, sol eats my 20x usage.

It used to feel unlimited, but not so much anymore. It's been a productive week, but I blew through my weekly limit with a reset.

So, I switched over to cursor and finally got a chance to try out grok 4.6 fast tier.

Here's my initial thoughts:

- grok 4.6 fast tier is incredible. work that would take 30-40 minutes with sol high fast is done in 5-7 minutes.

- usage limits are not the same as an OpenAI sub. we all know that the OpenAI limits are very generous. perhaps I need to try grok 4.6 without fast tier enabled.

- grok doesn't try to find as many edge cases as sol, and it doesn't over engineer. sometimes little things can get missed on the first try. this could be thought of as either a pro or con depending on the task.

- my adhd likes the feeling of staying in flow, and grok 4.6 just feels more fun to use.

- the output formatting is kind of trash. this is a little thing, but codex does are really good job of formatting the outputs with bullets and sections. I could easily solve this with some instruction or a skill.

- I ran into a few tool call errors when trying out computer use requests. codex has this figured out, and it works extremely well. I also like how sol will not hesitate to use computer use to solve an issue if it runs into a roadblock.

- just like grok 4.5, the 4.6 model understands my intent very well. after working with sol for a while on a project, I'm finding a lot of things that were over engineered, and grok is helping to make things more efficient.

overall, it's a great model. over the last few weeks it finally feels like we have choices for picking models that can perform extremely well and can be trusted.

[Logan @itsLoganAbell]

@coreyganim Just another reminder we’re so early with AI lol. Granted you’re working on another sales op right now, but the train hasn’t even left the station.

That line made me chuckle. Thanks for sharing and great work, keep it up.

[Logan @itsLoganAbell]

@Persist_Change How are you framing the engagement? For $25 - $50 a month are you just telling them about a specific tool they could use?

[Logan @itsLoganAbell]

@Teknium @pritmish Do you have Hermes doing the coding or do you have it connected to another harness? I saw the other day you mentioned you have just a single profile, is that still true? Or are you doing some style of profile orchestration?

[Logan @itsLoganAbell]

when the models stop being subsidized, the local LLM crowd is going to be screaming "I told you so".

[Logan @itsLoganAbell]

@IQReactorAI solid little home lab setup.

[Logan @itsLoganAbell]

what model is equivalent to gpt 5.6 sol but is actually fast and enjoyable to work with? it's hard staying in flow when the model over engineers everything I give it. grok 4.5 felt like we were almost there.

[Logan @itsLoganAbell]

@codylschuldt @NetworkChuck do you think having lots of profiles is better than just one profile with skills or one profile with topics? and if so, do you think its a personal preference thing or it's vastly superior?

[Logan @itsLoganAbell]

Found a new GPTism in 5.6.

“This is an unusually good idea.”

Been seeing this a lot. Of course it’s a good idea that’s why I’m asking you to do it.

[Logan @itsLoganAbell]

Remember that your ideal client is not on X. They have no idea what an 'agent' is. They have heard about AI, but it's just this thing that's going to take everyone's job.

This is what their mainstream news algorithm tells them on their phone.

I am trying to remember that X is a bubble. What my timeline shows me, and likely yours as well if I am seeing your content, is what 0.1% of people see about AI.

I am trying to frame my services around the fully managed concept. I will work with the client, guide them, setup the workflows for them, teach them how to use their agent effectively, and help find where the bottlenecks in their business exist that would benefit most from some sort of AI automation.

The other day I set a requirement for anything I would put in front of a client. "Would my wife use this?". It's a gentle reminder to myself that most people are not technical and don't have the want, the time, or the understanding on how to set things up.

[Logan @itsLoganAbell]

@coreyganim I love the grind man. Tough crowd though, that is a hard vertical to sell too. The reality is the type of person in this industry could care less about anything IT related. You may have better luck targeting larger companies that have more overhead.