Daniel Gallagher

@NoahKingJr Pen and paper

What’s coming after AI?

@ecom_cork Is that not the same person from two different angles?

Why do all these guys start looking the same ?

Do any airports have gyms? Flight just delayed an hour and would be the perfect way to kill the extra time

If you had to pick one, which moves the needle more on AI output quality: A better prompt, or better contextual data? Where do you land?

@vivoplt True but what's the alternative? Spend $$$ training your own model on a problem that may not even have customers? I think it's a good way to validate a solution before overcommitting

90% of "AI startups" are just: Take user input Send to OpenAI API Display response Charge $29/month

@breallstrong The Tier 2 jump is underrated. Most people stay at Tier 1 forever because building context feels like wasted time. What does Tier 2 look like for you in practice: context docs, custom instructions, MCP tools?

Reaching a point where I've finally abstracted one level higher in my AI usage. Up until recently, I was a "tier 1" AI user. Prompt, and re-prompt until I got my desired outcome. Now, I would consider myself "Tier 2." I'm building out context that make every single prompt more effective. Excited to keep progressing.

@bryandstarck @AnthropicAI What caused the ban?

Well, getting my company's @AnthropicAI account banned randomly for no apparent reason was not on my bingo card for this morning 🙃 Good reminder of why you can't become overly dependent on any of these individual providers...

Zurich has some beautiful run spots

@drewfallon12 How did you know it was a claude output?

just paid an agency $6k for a deck and got a claude output 😭

Had an audit call with a new client last week. They came in wanting automations. Looked at their setup and the real issue was underneath: data scattered across tools that don't talk to each other. It's holding together now because they're small and someone's manually keeping track, but that breaks the moment they grow. Ended up reshaping the project: a couple of automations for quick wins, but most of the work is fixing the data architecture first. Otherwise we'd just be automating on top of a mess.

@maxwellcopy What're you looking to build?

Yo I need some AI wizards to help us build some more tools internally. Who got me?

Before you build any AI solution, restructure the data it's pulling from. Most projects die here. People point a model at their Notion, Google Drive, and Airtable, get garbage outputs, and blame the model. What actually works: spend the first week in the data. Consolidate sources. Kill outdated docs. Add the context the needed to reason well — who does this apply to, what problem does it fix, what area is it relevant to. Fix the groundwork first.

@eptwts Grok did not do a good job lol

@danjgalla yea, i didn't expect that

gpt-image-2 has officially enabled fake MRR screenshots lmfao

@maxwellcopy How do you picture the manager interacting with them? One interface that lets them sequence jobs from a single spot, or a dedicated interface per area that goes deep on that domain? (design, strat, etc)

Boohoo figure it the fuck out or get left behind. The future of our agency will be one generalist manager (who knows a bit about everything) managing specialist agents. > A copywriting agent > A design agent > A strategy agent > An analysis agent All working under them… Their whole job is managing those agents, double checking, prompting, and keeping that human touch on the output. The tough part is that I only see founders trying to build this out right now. Employees aren't building their own agents, and I don't get why… If you get good at building agents or training agents, you'll be safe because you're going to be an agentic operator no matter what… You just need to know everything at a base level so you can double-check the agent. For us, what we're seeing inside the agency is that copy is trained well, and humans still oversee it. AI email design is mid because every tool right now sucks (won’t suck for long) Our designers are still the ones building emails, but we generate imagery and touch up photos with AI in about 50% of our designs. Lock in and figure it out now.

The difference between a useful AI agent and another generic chatbot is what's behind it. Designing a lead qualification agent for a client right now. We could just give it a script in its system prompt and let it run. Instead we structured their databases first so it has rich data to pull from, now the agent will be plugged into their live data, documentation, and lead conversation history. Will help qualify leads 24/7 when they're not online, follow their SOPs, and remember every conversation per lead

@JamesonCamp I see nothing wrong here

They should have never let me grow up and be in charge of my own nutrition

@MichaelRathman7 Saucony Ride 15’s

shill me your favorite running shoes plz

@apollonator3000 Optimise for the minimum number of operations required to get the job done

a lot of people in the AI space try dive deep into building AI tools & workflows before they even understand... - how LLM's work (diff models, context windows, token economy) - prompt engineering - context engineering - data manipulation i think building any flow w/ LLM calls is very stupid until you have a good understanding of these things you might find a way of performing a task, but it'll almost always be less efficient than it can be

@levelsio Definitely for flights AirPods run out of battery way too quick But they’re great for exercising

New trend spotted