Sam Altman shared OpenAI's current plan. It has three layers: build an automated AI researcher, use it to accelerate science and productivity, then give every person access to personal AGI.

The most important marker is OpenAI's internal belief that by March 2028, a significant fraction of its research could be done by AI systems working alongside researchers. That does not mean OpenAI already has a finished automated researcher. But it does show that OpenAI is making "AI that helps build better AI" part of its public roadmap.

The real story is the loop OpenAI is describing: use AI to speed up research, turn that speed into productivity and scientific progress, then eventually translate it into personal AGI access. If that loop becomes real, the advantage of frontier labs will depend less on any single model release and more on who can organize the strongest research machine.

SpaceX says Elon Musk will give a technical update on its ability to manufacture, launch, and operate AI satellites at scale. The way SpaceX is framing the story moves AI infrastructure beyond traditional data centers and toward something closer to a space data center.

As AI demand keeps running into power, chips, data centers, and deployment speed, SpaceX is putting compute inside a different frame: launch capacity, satellite manufacturing, and off-planet infrastructure operations.

Reuters reports that OpenAI has confidentially filed for a U.S. IPO after Anthropic. Reuters also says OpenAI could be targeting a valuation of up to $1 trillion, while SpaceX is also pursuing a very large listing.

That puts three AI-linked giants in front of the same question: can public markets absorb multiple trillion-dollar companies at once, especially when growth still depends on huge compute spending and long-term margins remain unproven?

OpenAI's IPO is not just a liquidity event. It is a test of how public investors will price AI when the hype has to sit next to compute costs, talent costs, infrastructure needs, and payback timelines. If frontier AI wants to keep scaling, public markets may become part of the AI stack.

OpenAI expands Codex into knowledge work: OpenAI says non-technical teams are using Codex to build internal apps, prepare executive materials, create dashboards, and turn creative briefs into work products. Codex is being positioned as a delegation surface for output, not just a developer tool.

Kimi Work turns the local desktop agent into a product category: Kimi.ai says Kimi Work is a local desktop AI agent that can run many agents in parallel and navigate the web through its WebBridge extension. If cloud agents are one path, local desktop swarms are another: closer to files, web sessions, and personal workflows.

OpenAI wants an evidence layer for AI's economic impact: OpenAI launched the Economic Research Exchange to support external research on how AI affects workers, firms, institutions, and the broader economy. As OpenAI moves closer to public markets, the question of how AI creates economic value will need more evidence and fewer anecdotes.

Meta AI's usage is rising, but retention is the real test: Meta AI has grown 2.5x in two months and could become the third-largest consumer AI app behind Gemini and ChatGPT, but that its 30-day retention is only 4.5%. Distribution can create usage spikes quickly; retention shows whether the product actually keeps users.

AI could create more software, not less: a16z quotes Benedict Evans arguing that AI will make new categories of software possible, which could increase competition. That pushes back against the simple idea that AI makes software less valuable; if software gets cheaper and easier to create, the market may expand.

Barbell strategy for the superhuman AI age: Ali Yahya suggests a barbell strategy: get very close to AI so you can learn, think, create, and work faster, while staying far enough away to preserve independent judgment, taste, and non-AI leverage. AI fluency is most valuable when it does not erase the human judgment layer.

Coding is only one part of engineering: Boris Cherny points out that engineering also includes debugging, operating services, scaling infrastructure, choosing what to optimize, capacity planning, talking to users, and product planning. Code generation is getting easier, but shipped engineering output still depends on the harder surrounding work.

If Codex and agent tools are expanding from coding into knowledge work, the best use case is not asking AI for one-off help all day. Build a menu of work you can delegate to AI, with context, boundaries, and review steps.

Here’s how you can build a first version that is useful enough to start with:

How to read the output:

Investors face a dilemma. When the S&P 500 finished its worst quarter since 2022 last month, diversifiers like bonds and bitcoin fell too.

Even with the turnaround in mid-April, analysts at Goldman Sachs and Vanguard have projected low-single-digit annualized returns from 2024-2034.

Bloomberg asked where experts would personally invest $100,000 for their March monthly edition.

One answer that surfaced for a second time? Art.

It's what billionaires like Bezos and the Rockefellers have privately used to diversify for decades.

Why?

Thanks to the world's premier art investing platform, now anyone can invest in works featuring legends like Banksy, Basquiat, and Picasso, without needing millions.

Shares in new offerings can sell quickly but...

Our subscribers skip their waitlist*

^{*According to Masterworks data. Investing involves risk. Past performance is not indicative of future returns. See important Reg A disclosures at masterworks.com/cd.}

Tommy argues that per-seat AI subscriptions may be priced far below the true cost of heavy usage, and that serious business use will eventually push products toward usage-based or capacity-based pricing. If that thesis is right, many AI tools will have to choose between higher prices, stricter limits, or longer burn.

Overlap quotes Pete Koomen saying good AI software is often very small: just enough code to let the model shine. It is a useful lens for builders: the moat may not be a huge feature suite, but context, workflow, timing, and a narrow surface where the model creates clear value.

Axios lays out four pressures on the AI business: costs are too high, payoff remains unclear, infrastructure demand is not always as strong as expected, and financing is getting more expensive. It is useful context for OpenAI's IPO because AI technology may still be powerful while the AI business gets tested on unit economics.

Axios frames Anthropic as an IPO story built around enterprise adoption, safety positioning, and direct competition with OpenAI. Read alongside OpenAI's filing, public markets will not only compare model capability; they will compare revenue quality, customer mix, compute cost, and governance story.