Your AI Coding Assistant Just Got a Lot More Expensive — Here’s Why
Something interesting is happening across the AI industry right now: the costs of using AI are becoming visible in a way they never were before. For years, flat monthly subscriptions hid what was actually happening under the hood. This week, that started to change — and developers, businesses, and everyday users are feeling the shock. Today’s stories, from surprise billing to AI guardrails to new workplace assistants, all circle the same question: who pays for AI, and what do we get in return?
GitHub Copilot switches to usage-based pricing — and developers are not happy
Usage-based pricing means you pay for exactly how much of a service you consume, rather than a flat monthly rate — think of how your electricity bill works compared to a Netflix subscription.
Starting June 1, GitHub Copilot — the popular AI coding assistant used by 4.7 million paid subscribers — quietly flipped that switch. Instead of a predictable monthly fee, users now spend “GitHub AI Credits,” a token-based currency where tokens (small chunks of text or code that AI models process) are consumed every time the AI reads or generates something. The more you ask Copilot to do, the more credits you burn. Ars Technica was among the first to report on the widespread user frustration, with Tech Times noting that some power users are seeing bills jump 10 to 50 times higher than before.
The reason this hits so hard is something called agentic use — where instead of just answering a single question, the AI autonomously runs through multiple steps to complete a task (like refactoring a whole codebase or writing tests for an entire project). Each of those steps costs tokens. A developer who used to pay a flat $19/month might now run through their entire monthly credit allowance in a single afternoon of heavy work.
For everyday users, this is a wake-up call about how AI pricing actually works. That affordable flat fee was always subsidizing something — and now companies are asking users to pay closer to the real cost. If you use Copilot lightly, you may barely notice. If you’ve been leaning on it heavily for complex tasks, it’s worth reviewing your settings and monthly credit limits before your next bill arrives.
“GitHub Copilot pricing shifted to token-based billing for 4.7 million paid subscribers”
Why this matters: This pricing shift signals a broader industry trend — as AI does more, it costs more, and flat subscriptions may not survive the age of agentic AI tools.
ZeroDrift raises $10M to put a safety net between AI and your employees
Compliance in a business context means following the rules — legal regulations, company policies, industry standards — that govern what a company can say or do.
A startup called ZeroDrift just raised $10 million to solve one of enterprise AI’s messiest problems: what happens when a company’s AI assistant says something it shouldn’t. Their product acts as a middleware layer (a filtering system that sits invisibly between two pieces of software) — in this case, between an AI model and the employee or customer reading its output. Before a response ever reaches a human, ZeroDrift’s system scans it, flags anything that could create a legal or compliance problem, and replaces it with something safer. TechCrunch has the full details.
This matters because large companies — banks, hospitals, law firms — operate under strict rules about what can be communicated to clients or staff. Current AI models, even excellent ones, occasionally produce responses that are confidently wrong, legally risky, or just off-brand. Rather than building a custom AI from scratch, ZeroDrift lets companies use powerful existing models while wrapping them in a compliance safety net.
For regular people, this kind of technology is part of why the AI assistant at your bank or insurance company will feel different from ChatGPT — it’s been filtered. Whether that makes it more trustworthy or just more cautious depends on who’s doing the filtering.
“A new AI compliance service sits between AI models and end users to flag and replace any messages that might present a compliance problem.”
Why this matters: As AI moves into regulated industries, the market for “AI guardrails” is only going to grow — and ZeroDrift’s funding suggests investors believe there’s serious money in keeping AI out of legal trouble.
Microsoft launches Scout, a personal AI assistant for your workday
Personal assistant AI refers to software designed to handle routine scheduling, email, and administrative tasks autonomously on your behalf — not just answering questions, but actually doing things.
Microsoft announced Scout this week, a new AI assistant baked directly into Microsoft 365 apps like Outlook and Teams. According to The Verge, Scout is inspired by OpenClaw — an agentic AI framework designed to let software complete multi-step tasks independently. In practice, Scout can draft emails, reschedule meetings, summarize long threads, and handle the kind of low-stakes administrative work that quietly consumes hours of most people’s workdays.
Microsoft describes Scout as “the first real personal assistant we’ve offered customers” — a notable claim given that Copilot has existed for a few years now. The distinction seems to be that Scout is more proactive: rather than waiting for you to ask a question, it monitors your calendar and inbox and takes action without being prompted.
For anyone who spends their day bouncing between email and meetings, Scout is the kind of tool that could genuinely reclaim time. The bigger question is trust — are you comfortable with software autonomously acting on your behalf, and what happens when it gets something wrong?
Why this matters: Microsoft Scout represents a meaningful shift from AI as a tool you use to AI as a colleague that works alongside you — and it will reach millions of Office users almost immediately.
Also happening in AI
Nvidia CEO Jensen Huang made headlines this week for publicly calling out business leaders who use AI as an excuse to reduce headcount, a rare moment of pushback from inside the tech industry itself. Meanwhile, Microsoft is doubling down on its AI-everywhere strategy with Project Solara, a new Android-based operating system designed specifically for AI-powered gadgets and agents. On the policy front, President Trump signed a narrower-than-expected executive order on AI oversight, establishing a voluntary framework after industry groups objected to stricter rules. And in a story that feels almost intentionally ironic, rocket startup Impulse Space raised $500 million — specifically to hire human engineers rather than replace them with AI. MIT Technology Review also published a useful practical guide this week on how small businesses can use AI for accounting, design, and market research without a big technical budget.
What to watch
The GitHub Copilot pricing backlash is a preview of a bigger conversation coming across the entire AI tools industry — flat subscriptions made AI feel affordable, but usage-based billing will reveal the true cost of agentic AI, and not everyone will like what they see. Watch for other developer tools and productivity platforms to follow GitHub’s lead over the next few months. The real question is whether companies will build smarter usage controls to help users manage spending, or simply let the bills speak for themselves.