Home
Apr 22, 2026
View All

Claude Code Pricing Confusion, GitHub Copilot Changes, and What Claude Design Can Actually Do

·1 underrepresented voice

The Short Version#

Two pricing stories dominated today: GitHub Copilot restructured its individual plans in ways Simon Willison flagged as meaningfully worse for solo developers, and Claude Code's pricing confusion created genuine uncertainty about whether the $100/month number floating around is real — spoiler: probably not, but it's complicated. Meanwhile, Lenny ran a hands-on test of Claude Design that's worth reading if you're deciding whether to add it to your AI toolkit.

GitHub — Changes to GitHub Copilot Individual Plans#

Source: https://github.blog/news-insights/company-news/changes-to-github-copilot-individual-plans/ Credibility: High (first-party GitHub announcement, corroborated by Simon Willison's analysis)

What happened: GitHub restructured its Copilot individual plans. Simon Willison flagged this today as worth paying attention to — the changes affect what's included at each tier and how individual developers access Copilot features. The specific restructuring details are at the linked GitHub blog post, but the signal here is that GitHub is actively repricing and repackaging individual access, which tends to precede either a push toward team/enterprise tiers or a recalibration of what "free" means for this category of tool.

Key capabilities:

  • Plan restructuring affects individual (non-enterprise) Copilot subscribers
  • Changes to what features are included at each pricing tier
  • Shifts in how individual developers access and pay for Copilot

Why it matters for PMs: This is the second developer AI tool pricing story this week (Claude Code being the other). When GitHub and Anthropic are both adjusting pricing structures simultaneously, it's worth asking what's driving it — whether that's margin pressure, a push to move developers onto team plans, or a signal that usage patterns are changing. If your team's dev tooling budget includes individual Copilot seats, check what changed before the next billing cycle. More broadly: individual AI tool pricing is in flux, and the direction seems to be toward higher costs or reduced free tiers.

Critical questions:

  • Are features moving from individual to team-only tiers, effectively forcing upgrades for collaborative use cases?
  • What's the net effect on solo developers vs. developers at companies with existing enterprise agreements?
  • Is this a precursor to GitHub consolidating individual plans and pushing harder toward Copilot for Business?
  • How does this interact with the free Copilot tier Microsoft announced for VS Code users?

Action you could take today: Pull up your team's current Copilot plan configuration and check whether the restructured tiers change your cost or feature access — especially if anyone on your team uses individual (not enterprise) plans.

Simon Willison — Claude Code Pricing: Is It Actually $100/Month?#

Source: https://simonwillison.net/2026/Apr/22/claude-code-confusion/#atom-everything Credibility: High (Simon Willison is a highly reliable analyst of AI developer tooling; this is his direct analysis)

What happened: A $100/month number for Claude Code has been circulating, and Simon Willison dug into whether it's accurate. His conclusion: probably not — "it's all very confusing." The confusion stems from how Claude Code pricing intersects with Claude Pro, Claude Max, and API usage, which creates a situation where the effective cost depends heavily on how you use it, what tier you're on, and whether usage overflows to API billing. This isn't a simple "$X/month" product right now.

Key technical details:

  • Claude Code pricing involves multiple overlapping plans (Pro, Max, API)
  • Usage-based overflow billing creates unpredictability for heavy users
  • The $100/month figure appears to apply to a specific high-usage configuration, not the baseline
  • Actual costs vary significantly based on usage patterns

Why it matters for PMs: Pricing confusion is a product problem, not just a comms problem. When your pricing is complex enough that a technically sophisticated observer like Willison needs to write a post explaining "it's confusing," that's a signal that the product isn't meeting basic customer expectations around cost predictability. This matters for PMs in two ways: (1) if you're evaluating Claude Code for your team, factor in that your actual spend may be hard to forecast, and (2) if you're building a product with complex pricing, this is a live case study in how pricing confusion creates friction even when the underlying product is strong.

Critical questions:

  • Does Anthropic have a known cost predictability problem across their products, or is Claude Code an edge case?
  • What's the churn signal when users can't predict monthly spend on a developer tool?
  • At what point does pricing complexity become a sales objection that outweighs capability advantages?
  • How does this compare to GitHub Copilot's pricing clarity (or lack thereof) for individual users?

Action you could take today: If you're on Claude Pro or Max and using Claude Code, run a quick audit of your last 30 days of actual usage costs vs. what you expected to pay. If you're evaluating it for your team, build a cost model with low/medium/high usage scenarios before committing.

Lenny Rachitsky — What Claude Design Is Actually Good For (and Why Figma Isn't Dead Yet)#

Source: https://www.lennysnewsletter.com/p/what-claude-design-is-actually-good Credibility: High (hands-on product test by a recognized PM practitioner with a track record of practical AI tool evaluation)

What happened: Lenny ran a real test of Claude Design — built a landing page, created slides, and did "an unhinged redesign" — to figure out what it's actually useful for. He also weighed in on whether GPT Image 2.0 changes how you'd approach brand creation. The framing matters: this isn't a "here's what the feature does" summary, it's a "here's where it helped and where it fell short" evaluation from someone who actually used it.

Key patterns:

  • Claude Design tested across landing page creation, slide design, and visual redesign tasks
  • GPT Image 2.0 compared as an alternative for brand asset creation
  • The "Figma isn't dead yet" framing suggests Claude Design has clear capability gaps vs. professional design tools
  • Practical assessment of when to use each tool vs. when to default to traditional workflows

Why it matters for PMs: PMs are increasingly using AI design tools to mock up concepts, build decks, and validate ideas faster. Knowing where Claude Design actually delivers (and where it doesn't) is more useful than knowing it exists. If the conclusion is that it's good for rough concepts and low-fidelity ideation but not production-ready design work, that's a clear workflow signal: use it for speed in early discovery, not as a replacement for design collaboration. The GPT Image 2.0 comparison is also worth noting — there are now multiple credible AI image/design tools, and they're differentiating on specific task types.

Critical questions:

  • What's the quality bar for "landing page" output — is it usable for actual user testing, or just for internal alignment?
  • How much prompt engineering is required to get useful output vs. just generating something that looks plausible?
  • Where specifically does it fall short vs. Figma — is it fidelity, interactivity, collaboration, or design system consistency?
  • Is GPT Image 2.0 better for static brand assets while Claude Design is better for layout/structure tasks?

Action you could take today: If you haven't tried Claude Design yet, give it one real task today — take an existing landing page concept or deck you've been meaning to mock up and run it through. The fastest way to calibrate a tool is to use it on something you already know how to evaluate.

OpenAI — Introducing OpenAI Privacy Filter#

Source: https://openai.com/index/introducing-openai-privacy-filter Credibility: High (first-party OpenAI announcement)

What happened: OpenAI shipped an open-weight model specifically for detecting and redacting personally identifiable information (PII) in text, claiming "state-of-the-art accuracy." This is a purpose-built model for a specific enterprise use case — PII detection — rather than a general-purpose capability. The open-weight release means teams can run it locally or integrate it into pipelines without sending data to OpenAI's API.

Key technical details:

  • Open-weight model (can be self-hosted)
  • Designed specifically for PII detection and redaction in text
  • Claims state-of-the-art accuracy for this task
  • Enterprise and privacy-sensitive deployment use case

Why it matters for PMs: PII handling is one of the top blockers for enterprise AI adoption — legal, compliance, and security teams routinely flag it as a reason to limit AI tool usage. An open-weight, self-hostable PII filter from OpenAI removes a genuine friction point in the "can we use AI on real customer data?" question. For PMs at fintech, healthcare, or any regulated industry building AI features on real user data, this is worth evaluating as a preprocessing step before data hits your LLM pipeline.

Critical questions:

  • How does it perform on domain-specific PII (financial account numbers, medical record identifiers) vs. generic PII (names, emails)?
  • What's the false positive rate — does it over-redact in ways that make downstream AI outputs less useful?
  • Does "state-of-the-art" hold up against specialized PII detection solutions already in market?
  • What are the licensing terms for the open-weight model in commercial products?

Action you could take today: If your product handles user data and you've been blocked from AI feature development due to PII concerns, flag this model to your security or compliance team as a potential preprocessing layer worth evaluating.

Quick Hits#

The Thread#

AI developer tool pricing is getting messier, not cleaner. This week: GitHub Copilot restructured individual plans (net worse for solo devs), Claude Code pricing confused a sophisticated analyst enough to write a dedicated post, and Anthropic is navigating the overlap between subscription tiers and API billing. The pattern isn't that these companies are gouging — it's that usage-based AI tooling is genuinely hard to price predictably, and companies haven't solved the UX problem of communicating cost to users. Expect more of this before it gets cleaner.

Sit With This#

Simon Willison's Claude Code post surfaces a real tension: a powerful developer tool with pricing complex enough that even experienced users can't predict their monthly spend. Pricing confusion creates churn even when the product is good.

For your product: Is there a feature or capability in your product where the value is clear but the cost (in money, time, or complexity) is hard for users to predict? What would it take to make that cost legible without simplifying it in ways that reduce flexibility?