Eric Barnard (@ericmbarnard) header
Eric Barnard (@ericmbarnard) profile picture

Packet Farmer. Product and Engineering at https://xitter.cf/t.co/4WPKnSJwVU

Champaign, IL
Joined April 2010
Photo by Eric Barnard (@ericmbarnard)Photo by Eric Barnard (@ericmbarnard)Photo by Eric Barnard (@ericmbarnard)Photo by Eric Barnard (@ericmbarnard)Photo by Eric Barnard (@ericmbarnard)Photo by Eric Barnard (@ericmbarnard)Photo by Eric Barnard (@ericmbarnard)Photo by Eric Barnard (@ericmbarnard)Photo by Eric Barnard (@ericmbarnard)
I'm out here trying to make grain marketing as simple as possible. If we can get rid of steering a tractor, we can make contracts, hedging, and insurance more straightforward. Advice is not my game - I build great tools. DMs are open. #AgTwitter
1
4
I support this
The new Big Ten Maps commercial, but set to the ‘Game of Thrones’ theme song:
103
Thats probably the nicest grain barge Ive ever seen
1
4
302
Made the 1000lb club today. 365 squat, 265 bench and 375 DL
3
9
691
This is literally me every night. I put the bed on the floor to make the roll off a little easier
A perfect escape. 😂
141
Im just gonna stand here and make sure this come-along cable doesnt snap
How to use a tire to easily remove a tree stump
2
173
Even if we get 10x better reasoning in the next wave of models, I see 2 major problems that will likely delay agents being real: cost and reliability. In the current prompt-in-text/data-out, costs are already brutal on frontier models to the point where product margins are razor thin compared to traditional SaaS. When it comes to agents, running the current frontier models in a loop is comical at current price points. It’s ngmi. At minimum, new middleware is going to be needed to properly select model size for the task(s) at hand, queue + batch process, and efficiently pass around state, files, and memory. Even if that’s all solved (which it’s not), there’s another elephant in room: reliability. Frontier model provider’s error rates are abysmal relative to other parts of the stack. This problem is already acute when all you’re doing is little chat or completion. But with agents, you’re effectively rolling the dice again and again and again. Reliability must be at nearly 100% for them to work and that’s so far from reality right now.
57
72
35
939
430,947
Exactly what we have @harves_iq … forgot to mention automated end-to-end tests that gate deploys and prevent regressions. Were not perfect but we rarely make the same mistake twice
“Deploy daily.” It’s a wonderful goal. But to safely deploy daily, you need a mature foundation: Automated build & tests No-downtime automated deploy Integrate daily (small deploys) Feature flags (hide incomplete work) Monitoring to detect issues Ability to quickly fix prod
1
238
Kickstarter is live for our new NES action platformer, The Storied Sword. Come help us reach our goal! kickstarter.com/projects/bit… Please help spread the word! Like and RT for a chance to win a prize pack of our previous NES games. (Bite the Chili and @0000jordan)
7
46
8
87
20,034
Pretty sure I can hear coyotes howling in the neighborhood… that or its other parents with small children.
1
41
They just reported the outage will last till morning. Im not sure what to do with my spouse this evening. We generally fall asleep in our chairs doom scrolling… but they just throttled our 4G data plan for trying to watch too many YT streams at once…
33
Hour 6 of @i3Broadband outage … and its getting tough. My kids are curled in balls on the floor sobbing bc all their toys only work w/ companion apps on the iPad. I showed them a basketball but they said they werent that desperate yet.
1
4
196
We ended up burning their Melissa & Doug toys to stay warm bc well … yeah and our wifi thermostat is stuck in a loop trying to display the current weather forecast
1
1
35
Just discovered this corner of the internet where people in Thailand are strapping a 12 valve Cummins diesel on to a small boat and racing. I have never been more inspired
714
4,873
1,076
56,173
6,827,144
Do NOT mock the database (or whatever is the primary data store). Majority of backend code is just data reads and writes. Biggest value in testing is knowing those code paths work. You can make those tests run in parallel and be fast with much less work than it takes to mock pg
"Thou shalt mock nothing [except third party services] in tests" - @alexdotjs Do you agree?
158
Browsers? This is not a new concept. Instead of manually crafting the buttons and textboxes AI will handle that … but devs likely just move up and focus on training AI to generate contextually correct UX
i wonder if the end state of all this is there’s just a single program instead of 10000s like we have today why have 2 apps that do 2 things if 1 app can do both a generate a UI that makes sense for each “work” would just be bringing physical resources into this program
168
Yep. Add: pray that other 2 kids sleep at least 6hrs
Day in the life of a small business guy: 4:30 - feed baby & pray he doesn’t puke - listen to Bible 5:30 - peloton & hope I don’t puke - listen to peloton pop music 6:30 - help get other kids up - listen to giggles and grumbles 7:00 - head to office 7:30 - set priorities for day and emails 9:00 - office team meeting for week 9:20 - rough draft training curriculum 11:00 - film “why choose us” video 1:00 - 50k uncomfortable conversation 3:00 - pick up daughter from school 4:00 - training curriculum & math games 4:30 - sales manager prep for week 5:25 - head home Post work routine sprint - Dinner, bed time, feed baby, talk with wife, head to bed Do it again
2
101
I raked the yard … in Feb 🫠
Midwesterners when there’s no snow, it’s too early to mow the grass and there isn’t football on
88