Did you actually try with other tools and see the cost differs? The pricing tables per same models between vendors seem very similar for me - don’t boiled down to harness co text bloat and also there I haven’t seen much diff
I literally am running Claude code, windsorf/devin and copilot in same machine and can’t say I see too much drift between enterprise usage plans cost
With the rise of @GitHubCopilot UBB and higher usage costs, teams and individual developers need to learn how to walk again.
For people coming from Claude, Codex, and other token or credit-based subscriptions, this is probably nothing new.
And don’t get me wrong, I myself am struggling both at work as a person in charge of cost and policy, and with my personal account(literally thinking twice before each session - and maybe that’s not a bad thing)
But for millions of developers who grew accustomed to (ab)using the PRU era (and I'm very much to blame with all my "use sub-agents, it's free" posts), this is going to require a mindset shift.
Not every problem is an Opus problem.
Not every task needs a swarm of agents.
Not every prompt deserves the most expensive model.
The funny thing is that constraints usually create better engineering habits.
And if we're already heading in that direction, there's one thing I'd ask:
The lower-tier models still leave a lot to be desired.
@burkeholland, we need the BEAST family back.
What are you doing to make this sustainable?
@jasmin_virdi@techgirl1908@GitHubCopilot That’s probably the progression
We spend days planning and iterating which is 100s of dollars worth of salary so paying a few dollars for a plan iteration sounds ok-ish, but as we got used to it costing cents I understand why it is hard to grok now
Now that we know how much @GitHubCopilot is gonna cost us, let's see what we are paying for :-)
New /chronicle feature landed in @code insiders
Use /chronicle:tips to get curated set of tips how to improve your token consumption and more and use /chronicle:standup to get list of recent sessions and summary of them
This is first step for local indexing and next iterations will add saving this to cloud so you can personally consume it or sharing it for all repo contributors to improve overall repo work by getting deep insights on all members work on the repo
Behind a feature flag: "github. copilot. chat.localIndex.enabled": true,
@Kphummingbird@GitHubCopilot@code@pierceboggan There’s more than we can chew - check the Cache Explorer
https://t.co/AbVR9M5D97
And if u wanna extract this data you can use OpenTelemtry and collect it from everyone in one place
New token controls dropping in @code insiders
You can now view your cached tokens in the Agent Debug Logs and drill down using a new Cache Explorer
View each turn, prompt signature(system, user, assistant, tool and drift), request components and cache performance stats and insights like "where the cache broke" etc
This is a great tool to learn how to effectively re-construct your prompting for better model and cost utilization
The GitHub Copilot app is now available for more people, good time to try it
I'm learning new things every day, for example the way it renders the plan with updates is absolutly beutiful!
The GitHub Copilot app technical preview now extends to all Copilot Pro, Pro+, Business, and Enterprise customers across Windows, macOS, and Linux.
https://t.co/buD0dt09QM
@JohannesVink@GitHubCopilot It is basically the new agent host protocol so behind the scenes it is the CLI
It’s a smart move of them exactly due to what u say - less runtime but they keep the surfaces so if u like cli use cli but if u like ide use ide with cli features full parity
A guy walks into a doctor.
Doctor: “I’m sorry, you’re going to die.”
Guy asks for a second opinion.
Doctor: “You’re ugly too.” 😅
@GitHubCopilot's rubber duck feature lets an agent ask another model family for a second opinion and review its plan, code, or tests before moving forward.
And it is now available in @code (it was already available in CLI and the new Github Desktop app)
@gsemetfr Fair callout - I mentioned it in the thread to be cautious about context manipulation tools as the results are not promised and it might cause more turns and more context bleed
So if anyone is onboarding these tools they should measure carefully
@Cubox_@GitHubCopilot TBH that boggles me as well, I really hope they open source it soon
Re:pricing - I’m afraid this is just the start of a price spike with all inference providers, but hope I’m wrong
🧵 The @GitHubCopilot Token Tax is Here.
Are You Ready for UBB?
Tomorrow, the rules of the game change. @GitHubCopilot is officially moving from the flat-rate Premium Request Units (PRUs) to Usage-Based Billing (UBB).
Here is a list of unfiltered options for you (and also untested ) to analyze your usage and start saving on your precious tokens.
A thread 👇
@Cubox_@GitHubCopilot That’s harsh but truth be told all tools are sooooo good I think you can’t go wrong
What did you find with Codex that was missing with Copilot?
@AhmedHamdy29189@GitHubCopilot@github@MicrosoftAI I think they have been experimenting with Raptor for quite a while, and it is exactly as you say
We have Build happening in the next two days so I'm sure a lot of news will flow
@AhmedHamdy29189@GitHubCopilot@github I'm pretty sure things will change now they move to UBB and there are rumors of MSFT models coming - and you can always use BYOK and bring your own models
That's true. There is a serious adoption problem.
Anthropic and OpenAI aren't pouring resources into FDEs, customer engineering, and adoption teams by accident.
AI adoption is death by a million paper cuts.
The interesting part is that in many cases the issue isn't the product capability itself.
The @Copilot Excel integration has become genuinely impressive over the last month or so. The Teams integration is also surprisingly good.
I don't have access to Cowork yet, so I'll reserve judgment there(although I’m dying to get access already).
The harsher truth is that low usage is often an organizational problem disguised as a product problem - and that’s the real
Problem for Satya and Co.
Successful AI adoption requires leadership, enablement, workflow redesign, technical champions, measurement, and continuous reinforcement.
Not every company is structurally prepared for that.
Buying licenses is easy.
Changing how people work is the hard part, which I personally deal with daily.
So getting CSO and CFO to sign off is not the end of it, it’s not even the start.
And you want a signal for true adoption? Just look in your feed and see how many people aren’t sleeping tonight to finish their quota before tmrw’s @GitHubCopilot move to UBB.
“Ex-Microsoft exec says the company blew it with Al, as it did with mobile”
"Not even 3% of paying Copilot users use it even when it's pre-deployed right in their faces”
The Microsoft 3% problem. See Word and Excel features.
Thanks for the feedback - that's one of the caveats - people are not aware of all the amazing tools
Happy I could have helped - and as I said - beware as there are still some rough edges they are carving out
If you want to share feedback comment here and I'll make sure it passes along