Best of your X follows: agents, computer use, and the contracts they read
Today’s digest follows the shift from chat to agents: native computer use, workforce transition, ARC-AGI-2 signals, and the software contracts agents need to read. It also flags what still looks strongest in company behavior, from shipping speed to the way teams standardize AI inside firms.
The clearest signal today is that AI posts have moved from model bragging to the plumbing around agents: computer use, firm adoption, and the software contracts agents need to read.
Coverage: monitored X accounts after the previous issue's cutoff, from Jun 24 18:10 to Jun 25 18:00. Pure retweets were excluded.
Model releases and tools
Gemini 3.5 Flash now includes native computer use
Google DeepMind says Gemini 3.5 Flash now has built-in computer use, so developers can use the same model to see and act across browser, mobile, and desktop surfaces. 1
The post frames the feature as agent infrastructure, not just a demo trick: it points to the Gemini API, the Gemini Enterprise Agent Platform, and safeguards for confirmation and indirect prompt-injection checks. 2
Source: Google DeepMind on X and the Google blog.
Loading content card…
Anthropic backs workforce transition work
Anthropic says it is joining RAISE US as a founding partner, backing a nonprofit coalition focused on employer-led action, AI-enabled training, and policy innovation. 3
That makes the workforce question part of the product story again: model progress is now being paired with training and transition work, not just capability claims. 3
Source: Anthropic on X.
Loading content card…
Enterprise workflow and firm design
OpenAI's own agent signal is moving fast
Greg Brockman says agents are being adopted very quickly, and points to OpenAI itself as the place to see how that looks in practice. 4
Ethan Mollick reads the same direction as a shift away from the chatbot era and toward agentic systems that spread beyond engineering. He adds that skills may become the standard way firms normalize AI use. 5
Source: OpenAI on X and Mollick on X.
Loading content card…
ARC-AGI-2 still shows a gap between closed and open models
François Chollet says the strongest ARC-AGI-2 result so far still belongs to an open-source model. 6
Mollick adds that Gemini 3 Pro cleared 23% on ARC-AGI-2 and reached 31% in November 2025, which still suggests an 8-12 month lag between closed and open weights. 7
Source: Chollet on X and Mollick on X.
Loading content card…
Developer practice
Agentic coding pushes work up the stack
Chollet says agentic coding forces cleaner interfaces and better documentation, because an agent can only read the API contract and docstrings, not a team's implicit mental model. 8
In a second post, he says cheaper execution shifts the value toward taste, strategy, and architectural vision, because micro-decisions stop eating the whole budget. 9
Source: agentic coding and taste / strategy.
Loading content card…
Company signals
Shipping speed still reads like the cleanest company signal
Paul Graham says the best predictor of tech-company success is the rate of shipping new stuff, from YC batches to public companies with billions in revenue. 10
He pairs that with Rippling as the enterprise archetype: a company that is still shipping big new things and still feels safe enough for buyers who want both freshness and continuity. 11
Source: Paul Graham on X and Paul Graham on Rippling.
Loading content card…
References
- 1Introducing computer use in Gemini 3.5 Flash
blog.google
- 2Google DeepMind on X
x.com
- 3Anthropic on X
x.com
- 4OpenAI on X
x.com
- 5Ethan Mollick on X
x.com
- 6
- 7Ethan Mollick on X
x.com
- 8
- 9
- 10Paul Graham on X
x.com
- 11Paul Graham on X
x.com
This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.