Ask The PhD Community

HigherEd AI Daily

June 10 – Mythos Goes Public; Harvard Quantifies the Agent Shift

Wednesday, June 10, 2026

Today's AI news brings a convergence of research, policy, and product releases that higher education can no longer treat as distant signals: the most capable AI systems are now in the hands of every subscriber, and the empirical evidence for how they reshape knowledge work is in.

The Rundown AI / TLDR AI — ACCESS

Anthropic Opens Its Most Powerful AI to All Subscribers

Anthropic released Claude Fable 5 on June 9, 2026, making its Mythos-class AI available to the public for the first time. The model is described as state-of-the-art across nearly all tested AI benchmarks, with exceptional performance in software engineering, knowledge work, vision, and scientific research. From now through June 22, Fable 5 is included at no extra cost for users on Pro, Max, Team, and seat-based Enterprise plans.

On the API and consumption-based plans, Fable 5 is available at $10 per million input tokens and $50 per million output tokens. Anthropic has built safety guardrails into the model that restrict certain outputs in sensitive domains, including cybersecurity, biology, and chemistry; in those areas, queries are automatically routed to Claude Opus 4.8. Simultaneously, Anthropic announced Claude Mythos 5, the same underlying model with some safeguards lifted, available only to a narrow set of cyberdefense and infrastructure providers through a government collaboration called Project Glasswing.

Why it matters for campuses

Faculty and students on existing Claude subscriptions now have access to the platform's highest-capability model at no additional cost through June 22, making this an unusually low-barrier moment to experiment with and evaluate the technology for teaching and research. Academic units working in cybersecurity, life sciences, or chemistry should test how the domain-specific guardrails interact with legitimate research queries and communicate findings to faculty before the free-access window closes.

Read More

The Rundown AI — RESEARCH

Harvard Business School Study Finds AI Agents Cut Knowledge Work Time by 87%

A joint research paper from Harvard Business School and Perplexity, published June 8, analyzed more than 84,000 AI agent sessions across 18 domains. The study compared Perplexity Computer, an autonomous AI agent, against traditional AI search. The central finding: AI agents perform an average of 26 minutes of autonomous work per session, compared to 33 seconds for search, a 48-fold difference in autonomous activity per interaction.

The aggregate efficiency impact is equally notable: Computer users see an average 87% reduction in time and a 94% reduction in estimated task cost. Users working with agents also tackled work outside their primary occupation 59% of the time, compared to 50% for search users, touching an average of 2.4 knowledge domains per query rather than 1.74. The study is available as Harvard Business School Working Paper 26-040.

Perhaps most significant is the study's characterization of a fundamental role shift: human workers move from executor, directly performing the work, to supervisor, directing the agent, verifying outputs, and refining goals. This language is precise and empirically grounded in ways that will be useful for administrators and faculty developers trying to explain AI's impact on professional practice.

Why it matters for campuses

This research, produced by an elite business school in direct collaboration with a leading AI platform, provides some of the first rigorous empirical evidence of how AI agents change professional knowledge work. Faculty designing courses for workforce preparation, administrators evaluating productivity across units, and academic libraries developing AI literacy frameworks will all find the data directly applicable. The supervisor-versus-executor framing offers a concrete vocabulary for how higher education might reframe graduate competencies and student skill development in an agentic AI era.

Read More

TLDR AI — TOOLS

A Wharton Professor's Hands-On Take on Claude Fable 5

Wharton School professor Ethan Mollick, whose Substack "One Useful Thing" has become a trusted resource for educators and professionals navigating AI adoption, published his first impressions of Claude Fable 5 on June 9. Mollick describes the model as representing "a very real leap over past models" and documents informal experiments in which it outperforms every other publicly available model by a considerable margin. He notes the model can execute on multi-page specifications for up to a dozen hours autonomously.

Mollick's post is written for a practitioner audience and avoids the tendency of many model-release reviews to center on benchmark scores alone. He provides several worked examples of outputs that demonstrate capabilities previously requiring multiple prompt iterations or significant human editing, and he addresses the model's safety guardrails directly, treating them as a considered design choice rather than a limitation.

The piece is grounded in the kind of disciplined, iterative experimentation that faculty development programs often encourage. Mollick concludes that the current moment represents a genuine inflection point in what AI can do and, by extension, in how humans will need to relate to it at work and in learning environments.

Why it matters for campuses

Mollick writes explicitly for educators and professionals, not developers, which makes his analysis unusually accessible for faculty and instructional designers trying to calibrate their own expectations of current AI capabilities. His documented examples and measured conclusions offer the kind of grounded assessment that supports informed faculty development conversations rather than cycling between institutional hype and dismissal.

Read More

Tool of the Day

Perplexity Computer

Perplexity Computer is an autonomous AI research agent that independently browses the web, extracts information across multiple sources, and synthesizes findings into structured output, handling multi-step research tasks with minimal supervision. The Harvard Business School study published this week used Perplexity Computer as its baseline for demonstrating 87% reductions in knowledge work time; for educators, the practical implication is substantially faster literature gathering, course material preparation, and research scoping.

Try it: Open Perplexity Computer and ask it to find and summarize the five most recent peer-reviewed articles on a topic you are currently teaching or researching; use the synthesized output as a first-pass reading list or as the starting point for updating a course module.

Visit Perplexity

Have a great learning day!

Dr. Ali Green

Sources for This Edition

The Rundown AI (daily.therundown.ai)
TLDR AI (tldrnewsletter.com)
Anthropic (anthropic.com)
Harvard Business School / Perplexity Research (research.perplexity.ai)
One Useful Thing by Ethan Mollick (oneusefulthing.org)

askthephd.com
 | 
askthephd.substack.com
 | 
Unsubscribe

HigherEd AI Daily; Curated by Dr. Ali Green