CMO & CTO @ Signos. Co-founder/CTO @ Mavenlink (scaled to $100M ARR upon exit). Writing as an operator on AI, SaaS, and what customers actually do
Salt Lake City, UT
Joined November 2008
- Tweets3K
- Following375
- Followers575
- Likes324
Pinned Tweet
We built the workflow we wished they had, not the workflow they wished they had
They kept using the spreadsheet
Customer validation is everything. Agentic coding makes the build cheap. It does not make the customer real. Validation is still the gate. The loop just got shorter
This is incredible. It’s the big boy version of having models help build and refine master prompts, skills, etc
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
The first time I saw my friend on E1M1 over our (probably) 14.4k modems at the time was an unimaginably cool experience as a teen and a huge upgrade over VT100 door games
We played plenty of DM, but a metric ton of coop. Crucial and underrated addition @ID_AA_Carmack @romero @SandyofCthulhu
I am hesitant to weigh in on discussions about 30+ year old details. I understand enough about memory in general, and mine in particular, to not have great confidence about it. I generally trust Romero’s memory more than mine — I was always a bit jealous of that capability.
In broad terms, multiplayer gaming was right up there with 3D gaming in my grand visions of what I wanted to accomplish. The early FPS games were very much “first person Gauntlet”, and Gauntlet had multiplayer.
I do clearly remember the first time I got two computers running Doom taking to each other. Moving around on one computer and seeing the character move on the other screen was sublime. And then, of course, I shot the other character. Every shared virtual environment inhabited by young men inevitably hosts deathmatch.
I suspect the confusion here may be around the fraction of people that originally bought Doom because of multiplayer. It was almost certainly well under 10%. Maybe under 1%. It wrecked some college networks, but most people didn’t have Ethernet cards or null modem cables. We probably did have some conversations correctly predicting that. Deathmatch was awesome, but most of our players never got to experience it.
Even four games later for Quake 3 Arena, being “deathmatch focused” was a huge internal battle, and not at all obviously viable.
There’s a very easy explanation for this: obesity makes sleep apnea about 5 times more likely
Retatrutide’s results mostly track the weight loss: higher doses cut 20–22% of body weight and substantially reduced breathing pauses during sleep, with up to half of people reaching mild or resolved apnea without excess daytime sleepiness
Extra fat around the airway is the main driver. Solve obesity, solve a lot of issues like this
Billboard Modern Rock Top 10 for the week of October 26, 1996. Not a bad playlist…
1. What I Got – Sublime
2. Novocaine for the Soul – Eels
3. Angels of the Silences – Counting Crows
4. Down – 311
5. I Was Wrong – Social Distortion
6. If It Makes You Happy – Sheryl Crow
7. Ready to Go – Republica
8. Muzzle – The Smashing Pumpkins
9. 6th Avenue Heartache – The Wallflowers
10. Burden in My Hand – Soundgarden
If we’ve just entered the Maclupa era, that means the delicious doritOS locOS macOS is right around the corner
I often talk about this a little differently. A business is built on 3 D’s:
- Demand
- Differentiation
- Defensibility
Defensibility == moat and there have always only been 2:
1. Data (information)
2. Network (connections)
Hardware is not a moat. Ask the folks at BlackBerry
Brisket and Ribs were the pacing strategy the whole time. Low and slow. Then, delicious output!
Impressive model drops today from @Meta
This is right, but not quite
The 3 roles with max leverage are (and always have been): Marketing, Product, Engineering
A founder is someone who’s great at one or more of those roles (not just marketing) and exceptional at finding talent for the other areas
Marketers: The next wave of founders will be marketers, distribution is the new moat.
PMs: Taste is the moat, and its easier than ever to build the wrong thing. Now is the age of the PM.
Eng: You need systems thinking and technical chops to solve hard problems and wield agents effectively at scale. Engineers have more leverage than ever.
I suspect good people in each discipline will remain valuable and there isn't one role poised to become unicorns en masse.
I do think its good to be skilled at more than 1 discipline, but I am not sure any one of these disciplines has a leg up over the others in that regard
I’ve been saying this for a while: if you ignore this platform due to your own hangups, but you’re in software/AI, you’re literally not plugged into what’s happening. Daily. Get over it for your own sake
@X is as alive today as Twitter was in the mid teens, perhaps more
I’m sure this one won’t make anyone angry. (But probably not here.)
daringfireball.net/2026/10/x…
I don’t know if this is an old axiom, but “the slowest service sets the pace”
I’ve been having a great time the past 2 weeks optimizing our @depotdev CI, @awscloud deployment, and @datadoghq monitoring stack with agents. Multiple x’s better. So easy to build, test, and measure!
Not sure why this has 800k impressions. It’s probably mostly dunks, but it’s an absolute masterclass in lean startup tactics
Let me explain:
There’s about a 1% chance there’s any TAM worth chasing here. Few people have mp3s, phones killed any need for the iPod, and you can easily just buy an eBay iPod with similar specs
But…insert Lloyd Christmas, “so you’re saying there’s a chance”
For the price of an AI produced video (free), you can test a market. Yeah, sure, ships in Q4 2027. Sounds great. Does anyone pony up? That’s a perfect fake door lean startup test
AI has made it free to test your ideas with highly polished artifacts like these nicely rendered views. There’s about a 1% chance this team has put any work into the actual execution: Vendors. Manufacturing. Hardware design. Everything.
That didn’t stop them from testing the concept. There’s a lesson in there…
Just don’t marry your ideas before you test them. That’s the only way you can lose
Introducing Sudo - A delightful music player for the songs you love
Reserve yours today:
sudo.music
Busy people pay for things that “just work” and they always will. It’s not easy to keep up with all the new capabilities and possibilities, even if using them is easy
The average business owner doesn’t have the time or the inclination to tinker with bots. It’s your opportunity
99% of people don't care to learn Grok Bot, Muse, Instinct, or Dots, even though using them is easier than ever.
getting agents to autonomously run a business and actually make money is an art and a skill in itself.
business owners 10 out of 10 times prefer to pay a high ticket fee upfront for someone to automate their entire operations into a business in a box.
the client texts the agent directly while the cloud computer stays completely invisible to them.
they will happily pay $25k to $50k upfront to deliver that within 90 days, plus a monthly retainer to maintain it.
baffles me that this is still the case, but its *absolutely* true
GTM teams take note:
I demo 20% of the product based on the expressed needs of the client
A harbor cruise makes them price the stuff they’ll never use
Enterprise buyers punish unused surface area or, worse yet, get overwhelmed with indecision
Stay brief, drive the convo
Saw this @ClaudeDevs post yesterday and added it to my account! You'll see it in Settings > Usage with a banner
What's the $250 for? @claudeai cloud sessions are included and free, but usually hit your usage. The credit gets prioritized for cloud sessions rather than your usage quota
Offload some work to the cloud while working in Claude Code or when you've hit your session cap! Happy coding
ALT claude code credits
I’m unreasonably excited for this and want to go in cold without spoilers. Here they are if you’re into that kind of thing
Wow, ok, Opus 5.5 today
Going to try a little Grok 4.7 vs Opus 5.5 experiment at some point
