👨💻 building the visual canvas for ai @guidotnow 🧪 hypothesizing @stratuslabs 💌 prev product @kit
Joined June 2011
- Tweets 3,195
- Following 1,086
- Followers 7,503
- Likes 11,060
297 Photos and videos
Pinned Tweet
Mar 15
You literally have to be unemployed to keep up with AI.
Which I am. So I read everything, track what matters, and toss the hype.
Starting tomorrow I'm sending that filter to your inbox every Monday.
dylanfeltus.com/monday
3
26
2,098
Every B2B software company is (or should be) building a "headless" version of their product. One that can be used by agents.
But "headless" doesn't mean "brainless".
You don't just wrap your existing APIs into an MCP server and call it a day.
The companies that succeed in the agentic era are those that take a thoughtful approach to *designing* an agentic user experience (AUX).
Yes, that will likely involve APIs, MCPs and CLIs.
But the difference will be in the *ergonomics* of the interface. We need to figure out *how* agents actually want to use our products/platforms. Because if all they wanted to do was use them like humans do, we have "computer use" for that.
I'm personally very excited about this new agentic world when it comes to B2B software.
HubSpot is all-in on building the #1 agentic customer platform.
Just posted this in a private Slack thread with the HubSpot exec team:
Being agentic is not just about agents running *on* our platform, it's about agents *running* our platform (being able to operate it). That's how you take AI from being a simple tool to a savvy teammate.
105
84
988
97,116
Apr 11
one of the differences I've noticed between opus and gpt:
when it creates a repo, gpt makes the readme about the scope of it's work, where opus makes the readme for the end user.
opus thinks about the final result and use case. gpt thinks too narrow on the task.
2
2
128
Apr 11
it’s a specific example with clues at the bigger problem.
you need to make gpt slow down and try to understand perspective before building.
71
Apr 7
amazing
if you're about to release a model that you know has the ability to reveal zerodays in every commonly used open source project you could delay release for a few years or spend another ten billion on alignment RL. or you could just secretly fix all the zerodays yourself first.
1
170
Apr 7
i feel like every new house in the future will be built to include its own LLM supercomputer AI agent hub.
screens throughout the house that blend in and look nice when not in use. available for generative UI as you live and need to visualize things.
1
1
129
Apr 7
local models are ready for this.
all we’re missing is the Tesla bots 😆
40
Apr 7
ok this works
Apr 7
setting up nanoclaw for a project. using the anthropic agent sdk.
does that still work with claude max sub or is that patched too
1
1
151
Apr 7
setting up nanoclaw for a project. using the anthropic agent sdk.
does that still work with claude max sub or is that patched too
1
1
320
Dylan Feltus retweeted
Apr 5
I'm calling it. AGI is already here – it's just not evenly distributed yet.
1,605
1,173
13,714
2,579,858
Apr 5
tomorrow's issue is focused on the recent changes by Anthropic.
including several solutions to help get your agents running in a way that doesn't break the bank. and doesn't drive you crazy in the process (cough cough gpt)
Apr 5
every monday i send out a curated list of resources for ai builders
join free. issue #4 out tmrw 👇
dylanfeltus.com/monday
143
Apr 5
curious about building your own model? the actual roadmap:
step 1: understand transformers
→ Karpathy "Let's build GPT" youtube.com/watch?v=kCc8FmEb…
→ "Attention Is All You Need" paper arxiv.org/abs/1706.03762
→ HF LLM course huggingface.co/learn/llm-cou…
→ course.fast.ai
step 2: fine-tune a model (fastest way to "your own")
→ Unsloth unsloth.ai — runs on consumer hardware
→ Axolotl github.com/axolotl-ai-cloud/…
→ HF TRL huggingface.co/docs/trl — SFT, DPO, GRPO
step 3: train from scratch
→ nanochat github.com/karpathy/nanochat — GPT-2 for ~$48
→ LitGPT github.com/Lightning-AI/litg…
→ datasets: FineWeb, RedPajama, The Pile (all on HF)
📚 more resources:
→ "Deep Learning" by Goodfellow, Bengio & Courville — the textbook deeplearningbook.org
→ "Build a Large Language Model (From Scratch)" by Sebastian Raschka — hands-on, code-first manning.com/books/build-a-la…
→ "Designing Machine Learning Systems" by Chip Huyen — the production/systems side
→ Karpathy's "Neural Networks: Zero to Hero" playlist youtube.com/playlist?list=PL…
→ 3Blue1Brown neural networks series youtube.com/playlist?list=PL…
1
3
130
Apr 5
Apr 5
every monday i send out a curated list of resources for ai builders
join free. issue #4 out tmrw 👇
dylanfeltus.com/monday
62
Apr 5
every monday i send out a curated list of resources for ai builders
join free. issue #4 out tmrw 👇
dylanfeltus.com/monday
259
Apr 5
agreed
Apr 4
my take is most b2b saas users don’t want a chat interface as their entry point.
they don’t want to ask for things, they want to see them
the real job as the builder is curation and deciding what data and tools show up first
not every platform needs to make you prompt your way to value
127
Apr 5
my agent preferred Resend for our projects so now I’m using Resend.
make your service an agents first pick.
Mar 30
Supabase: 1M → 4.5M developers in 12 months.
Resend: 0 → 500K users, beating a Twilio-backed incumbent.
Neither had a traditional growth team running this.
An AI agent picked both of them. Millions of times. Autonomously.
I wrote about how to build agent flywheels here: x.com/lavanyaai/status/20375…
1
1
117
Dylan Feltus retweeted
Apr 4
If you don’t want anyone to be able to lock you out (or price you out) of super intelligence you need to take running local models seriously
Apr 3
woke up and my mentions are full of these
Both me and @davemorin tried to talk sense into Anthropic, best we managed was delaying this for a week.
Funny how timings match up, first they copy some popular features into their closed harness, then they lock out open source.
26
12
255
37,140
Apr 4
a DIY project I started tinkering with is a local LLM running on OrangePi that works offline with solar power.
I already put together the software side of it. Going to connect the hardware and test it soon.
A silly project I hope we never need 😅
github.com/dylanfeltus/survi…
local LLM's would be insanely useful in an event that cuts us off from the internet, for example a world war...
would give you unlimited access to medical & survival info
i can't think of a single argument against getting a local LLM off-grid solar panel battery setup together
there's lowkey a million dollar product idea here
2
5
219
Apr 4
A good way to tinker with small local models too. That knowledge is going to continue to be more and more important imo
1
58