building ai agents for companies @ https://xitter.cf/t.co/WOtsAGQqAO
North Carolina, USA
Joined November 2011
- Tweets785
- Following999
- Followers594
- Likes5.7K
Pinned Tweet
Here's my lessons learned while building a product around AI agents for the past year: medium.com/@cpdough/building…
Instinct did a great job with the imessage integration and the emoji reactions and stuff but it just doesn't work to do anything real
“at this point, saying that AI will cure cancer is more a cliche than it is inspiring, and most people think it is deceptive. The thing that will work is *actually curing cancer*.”
- @DarioAmodei
Very much agree with this point holistically. CEOs don’t care at all when you tell them what AI will do for them, they only care once it starts doing that thing
"Two files I destroyed: the [....] feasibility doc and the [...] light proposal. Both untracked, so not recoverable from git, and there are no APFS snapshots. My fault: I put the rms in the same script as the cps without set -e"
Opus 5 keeps you on your toes man
if your agent configs are in a database you can give claude sql access to it and build an agent learning loop skill where it can update the config based on user interactions and then you don't have to work anymore
their October 2025 update said "Microsoft continues to have exclusive IP rights and Azure API exclusivity until Artificial General Intelligence (AGI)" sooooo... AGI now?
me, all day: "are you saying this is 100% safe to ship? have you considered all possible unintended consequences of this change?"
Random comments overheard at Augusta on Thursday:
- “Buyout beta just sucks man! It just sucks”
- (at 12pm) “it’s too early for you to have a beer at The Masters? What’s wrong with you?”
- Ludvig from the trees to his caddy: “don’t think we need the numbers for this shot man”
- Tommy after a bad 3rd on 13: “FOOKIN ‘ELL”
- After Akshay chunked it on 3: “Akshay I have that shot too!”
- Bathroom attendant: “keep it moving! 2 shakes maximum, any more you’re playing with yourself”… patron: “clearly you aren’t 50 yet, it takes a lot more than that”
- patron: “wish I brought my rangefinder”
- “I’m not going to say this loudly but this chicken biscuit is not that good”
The last year of AI advancements feels like Anthropic started with 3.5 Sonnet and tried to improve it at runtime and OpenAI started with gpt-4o and tried to improve it at runtime. The only progress made is that Anthropic just built a better original model so it looks a little bit better today but otherwise what are we doing?
The plateau is here
Sam Altman: "I bet there is another new architecture to find"
Sam Altman believes we are on the verge of discovering a new underlying architecture that will be as big of a leap forward as Transformers were over LSTMs.
He noted that we finally have AI models that are smart enough to help conduct this level of research (GPT 5.4 and above 👀) His direct advice to builders looking for the next major leap is to look for a "mega breakthrough" and use current models to help them find it.
claude: "Want to build this now or save it for a follow-up? We've shipped a lot today already."
that pre-AI training data about engineering velocity feels pretty silly now
“Every product will need an interface built for them” sounds a lot like an API?
Agents are becoming the primary users of software.
Every product will need an interface built for them. That's MCP. It's no longer a question of if you'll build an MCP server, it's when. Companies that don't provide one will get skipped entirely.
We're building @manufact (formerly @mcpuse) for this - full stack open source tooling to build MCP servers and the cloud to deploy and manage them.
Started with mcp-use in a bedroom in Zurich, now 5M+ downloads and $6.3M raised to make something agents want! Thanks @MichaelFNunez and @VentureBeat for telling our story
venturebeat.com/infrastructu…
working on a /feedback Skill for my ClaudeBro so had it go do research on entrepreneurs & books I felt like I learned from... I think it did pretty well:
The big 5 from the research:
1. No shit sandwiches (@bhorowitz) - don't buffer criticism with fake praise
2. Ruinous Empathy is the #1 failure mode (@kimballscott) - being too nice is worse than being too harsh
3. Pain + Reflection = Progress (@RayDalio) - discomfort IS the signal
4. Specificity or nothing (@rabois) - "you were unclear" is useless; "you didn't address the three most likely objections" is actionable
5. Pattern analysis > incident analysis (exec coaching) - one instance is noise, three instances is a blind spot
