There's a new type of AI model worth paying attention to, and it's basically just a really fast decision maker.
TypeSafe AI's JEV dropped this week. They're throwing a bunch of technical terms at it, but what it does is make decisions quickly. Click this, scroll here, is this worth looking at or not.
A ton of the workflow automations we build are just trying to get LLMs to make those quick decisions. Frontier LLMs are intelligent, but they're slow. Models like this are going to drop in there, speed things up and bring costs way down.
I'm starting to see a lot of the leading engineers running a two system setup.
Their AI coding agents already run on a separate sandboxed machine, away from passwords and private company stuff. That machine is the workhorse now, so the laptop doesn't need to be a souped-up MacBook Pro. People are looking at a MacBook Air or lighter Windows laptops instead.
It's a real shift in how spend works. Asking for two machines when companies are already paying for tokens is probably a harder ask. I'd personally be a fan of budgeting for both, but I'm curious how companies are handling it.
What's truly different this time is how fast the models are getting better compared to how fast humans can learn new things.
Before, tools got smarter about as fast as the people building them did. Now there's this divergence.
So split the work into two camps. There's the truly strategic stuff that can't easily be taught to a model. That's where you train your team. Then there's the stuff a model doesn't do great right now but will excel at within a few months or a year. It doesn't make sense to train your current team on that. I think those resourcing decisions will decide how fast you can move going into 2027.
One of the most common mistakes I see is teams getting an enterprise AI plan, handing out the chatbot, and locking down all the scary stuff.
Which seems really smart. But the chatbot is the thinking and talking part. You're basically saying it can think and talk, but it can't do anything.
The most valuable part is the doing: writing code, clicking around in apps, the grunt work. You do need the right compliance and security in place first. But once that's sorted, people start seeing their grunt work melt away. It's an easy trap to get stalled out in.
The era of $7,000 MacBooks for everybody on the team is kind of coming to an end.
A lot of the custom software people are building with AI coding agents, even non-technical founders and teammates, is going to run in the browser. It's basically a Chrome web page that feels like another app. So your computer just needs to be a screen that connects to wherever those apps actually run, whether that's computers at your office or something like AWS
That's why the Google Books look like a really good option. They're lightweight, cheap and easy to replace. And honestly, they're a lot less stressful to carry around!
๐ฅ Source: Cedlom โ "Did Google COOK with the Googlebook? ๐"
๐ https://t.co/ui3THt6K4j
Quick reminder: the subsidized Max and Pro accounts from OpenAI and Anthropic still give you a huge amount of compute for the money.
You're essentially paying 10 to 25 percent of what the same usage would cost at API pricing. Your company's enterprise plan doesn't get you that. You need the personal accounts.
For a while Anthropic blocked them on tools like OpenClaw and Hermes, so everyone went back to the API. They reversed that a couple months ago, and a lot of people just never came back. You can still wire them in and have them automating marketing flows, research, or building the little tools you're paying monthly subscriptions for.
We are honored to steward the future of the Indie mission!
2C will continue to offer expert Design & Development services, business as usual.
In the months to come, we will iterate on our learnings & ideas for the next version of future of work.
Stay tuned Indies!
We are excited to announce that Indie has been acquired by @twos_complement!
Huge thanks to the community for such a fun journey filled with ideas and creativity.
This concludes a wild first experiment, and it certainly won't be our last!
Read more: https://t.co/yPI72vIyKa
That app just acquired for $10B. Basic tech.
That company that just raised $100M. Basic tech.
Tech that has been around for a while is boring. But it's also:
๐ชจ Survived enough cycles to be ROCK SOLID
๐ฅ Adopted by enough devs to be EASY TO HIRE FOR
๐ Penetrated enough community to have TONS OF ADDONS
Devs like chasing the shiny new things. โจ It's fun. โจ
When you are building a competitive business though, you need something that can scale, and benefits the bottom line.
Of course, you need your devs to know modern tech and use the right tools for the job, but your core stack should be as popular, boring, and basic as possible.
It's your competitive edge.
Save your creativity and fun factor for the novel parts of your product.
"Why on earth are we doing it this way?" said the manager.
"Because that's how the tool is sUpPoSeD tO bE UsEd!"
Tools should be working FOR YOU, not the other way around.
We've seen teams spend 30 minutes formatting reports because "that's how Jira works" when a simple spreadsheet would take 5 minutes.
Many of the slowdowns we see stem from diligent teammates trying to master the tool they have been assigned, but eventually becoming a slave to the tool itself.
โจ First the tool seems like a golden solution to a problem.
๐ง The team integrates the tool into their workflow.
๐ Finally, over time, the team starts doing things a certain way to make the tool work, and forgets why it was added in the first place.
Teams need to regularly revisit each tool in their stack and ask if it's still working for them.
If it's not, ditch it.
Your teammates are wired to solve problems and add the right tools to your process and YOU as a founder need to constantly reset the tools that have taken control of your team.
Building everything from scratch doesn't make you Netflix.
It makes you broke.
It's not even about cost though, it's about where your team spends their mental energy and dozens of small side effects.
When you choose managed services, you get:
โ one less thing to manage in your CI/CD process
โ one less thing to manage security updates for
โ one less thing to maintain documentation for
โ one less thing to onboard new teammates into
โ one less thing to provide customer support for
โ one less thing to keep up to date with best practices
โ one less thing to hold architecture in your mind for
โ one less thing to debug when it's not working
โ one less thing to design for
Each "one less thing" frees your team up to focus on what actually moves the needle for your business.