Everything happening in AI dev. Condensed into 5 stories.
Everything happening in AI dev. Condensed into 5 stories.
In this roundup you get the five most critical AI development updates—from Sequoia’s AGI claim and OpenAI’s code automation, to GitHub Copilot’s new token billing, Anthropic’s hidden cost rise, and a US influence campaign tied to OpenAI.
Everything happening in AI dev. Condensed into 5 stories.
The five stories that matter most across AI models, developer tools, pricing shifts, and industry politics. One source. No filler.
Apr 29, 2026
1. Sequoia calls it AGI
At AI Ascent 2026, Sequoia Capital made a declaration that sent ripples through the industry: long-horizon agents are functionally AGI, and they have arrived. The announcement came from Sequoia partners Pat Grady, Sonya Huang, and Konstantine Buhler, who argued that the current AI wave is categorically different from all prior technology cycles.
The claim did not stand alone on stage. OpenAI president Greg Brockman appeared at the same event and stated that roughly 80% of code at OpenAI is now written by AI. His framing positioned human attention as the new scarce resource, not model capability.
Not everyone accepted the narrative. Cognitive scientist Gary Marcus publicly called the broader AGI claims a "trillion-dollar delusion," pushing back on the scale of capital being deployed around assertions that cannot be independently verified.
Hint: Brockman's 80% figure is self-reported and not independently verified. Treat it as a directional signal, not a benchmark to copy-paste into your own productivity claims.
The ultimate shortcut to flawless AI results
Stop wasting time guessing prompts. Get consistent, professional AI results right from the first try, every time.
GitHub announced that all Copilot plans will move to token-based, usage-metered billing on June 1. The new unit is called GitHub AI Credits, priced according to token consumption across input, output, and cached tokens. The change affects every plan tier, from individual Pro accounts to enterprise seats.
The underlying reason is straightforward. GitHub's own documentation acknowledged that a quick chat question and a multi-hour autonomous coding session previously cost the subscriber the same amount, while GitHub absorbed the difference. With agentic workloads becoming the default, that model is no longer sustainable.
The sharpest impact lands on subscribers using premium models. The multiplier for Claude Opus 4.7 rises from 7.5x to 27x for annual plan subscribers on June 1. Sign-ups for Copilot Pro, Pro+, and Student plans have been paused ahead of the transition, and Opus models have been removed from the base $10 Pro plan.
Plan
Monthly price
AI Credits included
Copilot Pro
$10
1,000
Copilot Pro+
$39
3,900
Copilot Business
$19 per user
1,900 per user
Copilot Enterprise
$39 per user
3,900 per user
Hint: Annual Copilot subscribers face the steepest multiplier increases on June 1 and cannot auto-renew at current terms. Check your renewal date now.
At Sequoia AI Ascent 2026, Andrej Karpathy declared vibe coding obsolete and introduced a replacement framework he calls agentic engineering. The term is deliberate. "Agentic" signals that the new default is not writing code directly, but orchestrating agents that do. "Engineering" signals that guiding those agents is a technical discipline with real depth, not casual conversation.
Karpathy identified December 2025 as his personal inflection point. In November, he was writing roughly 80% of his own code and delegating the rest to AI. By December, that ratio had fully inverted. He delegated 80% to agents and found himself trusting the output without correction.
The framework positions the developer as a supervisor rather than a writer. The role shifts toward defining goals, reviewing output, catching failure modes, and maintaining architectural judgment. Karpathy's sharpest line from the talk captures the tension: "You can outsource your thinking, but not your understanding."
Hint: Karpathy published a full transcript of the talk at karpathy.bearblog.dev, optimized for LLM ingestion. Worth feeding directly to your agent setup.
Anthropic launched Claude Opus 4.7 with several headline improvements alongside an unchanged rate card. The model posts a 12-point gain on CursorBench over Opus 4.6, supports high-resolution image input up to 2,576 pixels, and introduces task budgets for long-running agents alongside a new high-effort level for complex coding tasks.
What changed without announcement is the tokenizer. The new tokenizer generates up to 35% more tokens for identical input text compared to the previous version. Because pricing is calculated per token, the cost of any given request rises in proportion, even though the listed rate is the same.
For developers running automated pipelines, the impact is invisible at a glance and real on every invoice. A pipeline calibrated on Opus 4.6 token counts will produce materially different costs on Opus 4.7 without any change in prompts or workflow logic. The key changes in the release break down as follows:
Task budgets - Long-running agents can no longer silently exhaust quotas without a defined ceiling.
High-effort level - A new setting for complex coding tasks that increases model thoroughness.
High-resolution image input - Support for images up to 2,576 pixels, expanding multimodal use cases.
New tokenizer - Produces up to 35% more tokens for the same input, raising effective costs across all usage.
Heads up: Audit every automated pipeline running on Opus 4.7 before your next billing cycle. Compare token counts against your 4.6 baseline to catch unplanned costs early.
A Wired investigation revealed a coordinated influence campaign operating under the name Build American AI. The organization is linked to a $100 million super PAC called Leading the Future, funded by executives affiliated with OpenAI, Palantir, and Andreessen Horowitz. The campaign paid social media influencers up to $5,000 per post to promote pro-American AI narratives and frame Chinese AI development as a national security threat.
Influencers received pre-written scripts and were instructed not to disclose the organization behind the spend. Posts were labeled as advertisements on platforms including TikTok and Instagram, but the funding source and policy agenda were not disclosed to audiences.
OpenAI issued a statement saying it has no corporate affiliation with Leading the Future or Build American AI and has provided no funding or support to either group. Palantir similarly denied involvement. Andreessen Horowitz did not respond to requests for comment at the time of the investigation.
Hint: The full funding trail is documented in Wired's primary investigation. Read the source before forming or sharing an opinion on this one.
