SpaceX is edging toward the 12th launch of its Starship rocket, which this time will see a new version of the massive vehicle head skyward.

The launch team has just completed a crucial test of the Starship’s first stage — the Super Heavy — at its Starbase facility in Boca Chica, Texas.

It shared a short video and some images (below) of the most powerful booster ever built with the message: “Cryoproof operations complete for the first time with a Super Heavy V3 booster. This multi-day campaign tested the booster’s redesigned propellant systems and its structural strength.”

No anomalies have been reported, setting the booster up nicely for a static fire test, hopefully in the coming days.

After that, attention will shift to the upper-stage Ship, which will also undergo preflight testing prior to being lifted atop the Super Heavy in preparation for launch.

SpaceX is targeting March for the 12th Starship test flight, according to a recent post on X by Elon Musk, the company’s CEO.

The updated Starship, known as Version 3, stands about a meter higher than its 123-meter-tall predecessor, and comes with more powerful engines and a series of design tweaks to enhance performance.

In an update on Sunday that surprised many, Musk said that SpaceX is pivoting its near-term focus to building a self-sustaining city on the moon, deprioritizing his previously stated ambition of building a city on Mars.

The CEO said the lunar plan is achievable in under 10 years, with the moon offering a more rapid path due to frequent launch windows (every 10 days vs. Mars’ 26-month alignments) and shorter trips (2 days vs. 6 months).

There’s also the added challenge of safely getting humans to the red planet, a feat that has yet to be achieved.

SpaceX’s Mars‑city efforts will continue as part of a long‑term plan, while the company will continue with its involvement in NASA’s Artemis lunar program using Starship.

Airport runways seem pretty simple, just another strip of asphalt or concrete not unlike the roads that our cars drive upon every day. We can even use these same highways as landing strips in a pinch, so you’d assume that the engineering for either isn’t that dissimilar. Of course, you can use a highway for an occasional emergency, but a runway that sees the largest and heaviest airplanes taxi, take off and land on a constant basis is a whole other challenge, as detailed in a recent [Practical Engineering] video and its transcript.

When you consider that an Airbus A380 the take-off weight is up to 550 ton, it’s quite clear what the challenge is for larger airports. Another major issue is that of friction, or lack thereof, as the speeds and kinetic energy behind it are so much higher. One only has to look at not only runway overruns but also when one skids off sideways due issues like hydroplaning and uneven friction. Keeping the surface of a runway as high-friction as possible and intact after hundreds of take-offs, tail-strikes and other events is no small feat.

Of course, the other part of runway engineering is for when things do go wrong and an airplane enters the runway safety areas, or overrun zones. This usually provides some flat and clear space where an airplane can safely bleed off its kinetic energy, with the collapsing surface of the EMAS technology being one of the best demonstrations of how this can be safely and dramatically shortened.

Another aspect not covered here that is part of these overrun zones are frangible structures, such as any localizer antennae of ILS, lighting, etc. Frangible here means that the structure easily collapses when a heavy airplane crashes into it without causing significant damage to the airplane.

It was the failure of such a design process that doomed the crew and passengers of Jeju Air Flight 2216 in December of 2024, when the airplane during an emergency belly landing skidded over the end of the runway. Although there was a lot of open space after the ILS localizer array with just a flimsy wall and further level fields, the ILS array’s base contained a poured concrete base on which the airplane effectively pulverized.

Looking for the most recent regular Connections answers? Click here for today’s Connections hints, as well as our daily answers and hints for The New York Times Mini Crossword, Wordle and Strands puzzles.

Today’s Connections: Sports Edition is a tough one. Some of the words in the purple category were completely new to me. If you’re struggling with today’s puzzle but still want to solve it, read on for hints and the answers.

Connections: Sports Edition is published by The Athletic, the subscription-based sports journalism site owned by The Times. It doesn’t appear in the NYT Games app, but it does in The Athletic’s own app. Or you can play it for free online.

Read more: NYT Connections: Sports Edition Puzzle Comes Out of Beta

Hints for today’s Connections: Sports Edition groups

Here are four hints for the groupings in today’s Connections: Sports Edition puzzle, ranked from the easiest yellow group to the tough (and sometimes bizarre) purple group.

Yellow group hint: Step up to the plate.

Green group hint: College division.

Blue group hint: Robert.

Purple group hint: Goaaaaaal!

Answers for today’s Connections: Sports Edition groups

Yellow group: Involved in an at-bat.

Green group: A Big 12 athlete.

Blue group: Bobs.

Purple group: Soccer slang.

Read more: Wordle Cheat Sheet: Here Are the Most Popular Letters Used in English Words

What are today’s Connections: Sports Edition answers?

The yellow words in today’s Connections

The theme is involved in an at-bat. The four answers are catcher, hitter, pitcher and umpire.

The green words in today’s Connections

The theme is a Big 12 athlete. The four answers are Cyclone, Jayhawk, Sun Devil and Ute.

The blue words in today’s Connections

The theme is Bobs. The four answers are Beamon, Costas, Feller and Uecker.

The purple words in today’s Connections

The theme is soccer slang. The four answers are howler, screamer, sitter and worldie

It’s the middle of February. It’s cold and wet across most of the US, and I am feeling especially reticent to lace up my running shoes and head out into the great outdoors, or even to the gym. Last year, I was in the same position – it was cold and dark – but I didn’t even have a gym membership, having moved house and being between gyms.

As someone who, like a golden retriever, needs a certain amount of exercise once a day, how did I cope? By equipping myself to better do workouts at home, of course, using sales events (like the current Presidents’ Day sales) to do it for less.

Until recently, the practice of building AI agents has been a bit like training a long-distance runner with a thirty-second memory.

Yes, you could give your AI models tools and instructions, but after a few dozen interactions — several laps around the track, to extend our running analogy — it would inevitably lose context and start hallucinating.

With OpenAI’s latest updates to its Responses API — the application programming interface that allows developers on OpenAI’s platform to access multiple agentic tools like web search and file search with a single call — the company is signaling that the era of the limited agent is waning.

The updates announced today include Server-side Compaction, Hosted Shell Containers, and implementing the new “Skills” standard for agents.

With these three major updates, OpenAI is effectively handing agents a permanent desk, a terminal, and a memory that doesn’t fade and should help agents evolve furhter into reliable, long-term digital workers.

Technology: overcoming ‘context amnesia’

The most significant technical hurdle for autonomous agents has always been the “clutter” of long-running tasks. Every time an agent calls a tool or runs a script, the conversation history grows.

Eventually, the model hits its token limit, and the developer is forced to truncate the history—often deleting the very “reasoning” the agent needs to finish the job.

OpenAI’s answer is Server-side Compaction. Unlike simple truncation, compaction allows agents to run for hours or even days.

Early data from e-commerce platform Triple Whale suggests this is a breakthrough in stability: their agent, Moby, successfully navigated a session involving 5 million tokens and 150 tool calls without a drop in accuracy.

In practical terms, this means the model can “summarize” its own past actions into a compressed state, keeping the essential context alive while clearing the noise. It transforms the model from a forgetful assistant into a persistent system process.

Managed cloud sandboxes

The introduction of the Shell Tool moves OpenAI into the realm of managed compute. Developers can now opt for container_auto, which provisions an OpenAI-hosted Debian 12 environment.

This isn’t just a code interpreter: it gives each agent its own full terminal environment pre-loaded with:

• Native execution environments including Python 3.11, Node.js 22, Java 17, Go 1.23, and Ruby 3.1.

• Persistent storage via /mnt/data, allowing agents to generate, save, and download artifacts.

• Networking capabilities that allow agents to reach out to the internet to install libraries or interact with third-party APIs.

The Hosted Shell and its persistent /mnt/data storage provide a managed environment where agents can perform complex data transformations using Python or Java without requiring the team to build and maintain custom ETL (Extract, Transform, Load) middleware for every AI project.

By leveraging these hosted containers, data engineers can implement high-performance data processing tasks while minimizing the “multiple responsibilities” that come with managing bespoke infrastructure, removing the overhead of building and securing their own sandboxes. OpenAI is essentially saying: “Give us the instructions; we’ll provide the computer.”

OpenAI’s Skills vs. Anthropic’s Skills

Both OpenAI and Anthropic now support “skills,” instructions for agents to run specific operations, and have converged on the same open standard — a SKILL.md (markdown) manifest with YAML frontmatter.

A skill built for either can theoretically be moved to VS Code, Cursor, or any other platform that adopts the specification

Indeed, the hit new open source AI agent OpenClaw adopted this exact SKILL.md manifest and folder-based packaging, allowing it to inherit a wealth of specialized procedural knowledge originally designed for Claude.

This architectural compatibility has fueled a community-driven “skills boom” on platforms like ClawHub, which now hosts over 3,000 community-built extensions ranging from smart home integrations to complex enterprise workflow automations.

This cross-pollination demonstrates that the “Skill” has become a portable, versioned asset rather than a vendor-locked feature. Because OpenClaw supports multiple models — including OpenAI’s GPT-5 series and local Llama instances — developers can now write a skill once and deploy it across a heterogeneous landscape of agents.

But the underlying strategies of OpenAI and Anthropic reveal divergent visions for the future of work.

OpenAI’s approach prioritizes a “programmable substrate” optimized for developer velocity. By bundling the shell, the memory, and the skills into the Responses API, they offer a “turnkey” experience for building complex agents rapidly.

Already, enterprise AI search startup Glean reported a jump in tool accuracy from 73% to 85% by using OpenAI’s Skills framework.

By pairing the open standard with its proprietary Responses API, the company provides a high-performance, turnkey substrate.

It isn’t just reading the skill; it is hosting it inside a managed Debian 12 shell, handling the networking policies, and applying server-side compaction to ensure the agent doesn’t lose its way during a five-million-token session. This is the “high-performance” choice for engineers who need to deploy long-running, autonomous workers without the overhead of building a bespoke execution environment.

Anthropic, meanwhile, has focused on the “expertise marketplace.” Their strength lies in a mature directory of pre-packaged partner playbooks from the likes of Atlassian, Figma, and Stripe.

Implications for enterprise technical decision-makers

For engineers focused on “rapid deployment and fine-tuning,” the combination of Server-side Compaction and Skills provides a massive productivity boost

Instead of building custom state management for every agent run, engineers can leverage built-in compaction to handle multi-hour tasks.

Skills allow for “packaged IP,” where specific fine-tuning or specialized procedural knowledge can be modularized and reused across different internal projects.

For those tasked with moving AI from a “chat box” into a production-grade workflow—OpenAI’s announcement marks the end of the “bespoke infrastructure” era.

Historically, orchestrating an agent required significant manual scaffolding: developers had to build custom state-management logic to handle long conversations and secure, ephemeral sandboxes to execute code.

The challenge is no longer “How do I give this agent a terminal?” but “Which skills are authorized for which users?” and “How do we audit the artifacts produced in the hosted filesystem?” OpenAI has provided the engine and the chassis; the orchestrator’s job is now to define the rules of the road.

For security operations (SecOps) managers, giving an AI model a shell and network access is a high-stakes evolution. OpenAI’s use of Domain Secrets and Org Allowlists provides a defense-in-depth strategy, ensuring that agents can call APIs without exposing raw credentials to the model’s context.

But as agents become easier to deploy via “Skills,” SecOps must be vigilant about “malicious skills” that could introduce prompt injection vulnerabilities or unauthorized data exfiltration paths.

How should enterprises decide?

OpenAI is no longer just selling a “brain” (the model); it is selling the “office” (the container), the “memory” (compaction), and the “training manual” (skills). For enterprise leaders, the choice is becoming clear:

Choose OpenAI’s Responses API if your agents require heavy-duty, stateful execution. If you need a managed cloud container that can run for hours and handle 5M+ tokens without context degradation, OpenAI’s integrated stack is the “High-Performance OS” for the agentskills.io standard.

Choose Anthropic if your strategy relies on immediate partner connectivity. If your workflow centers on existing, pre-packaged integrations from a wide directory of third-party vendors, Anthropic’s mature ecosystem provides a more “plug-and-play” experience for the same open standard.

Ultimately, this convergence signals that AI has moved out of the “walled garden” era. By standardizing on agentskills.io, the industry is turning “prompt spaghetti” into a shared, versioned, and truly portable architecture for the future of digital work.

Update Feb. 10, 6:52 pm ET: this article has since been updated to correct errors in an earlier version regarding the portability of OpenAI’s Skills compared to Anthropic’s. We regret the errors.

It’s certainly easy to feel jaded by the modern gaming space.

Great games continue to release apace; this month along has already seen bangers in Nioh 3 and Romeo is a Dead Man. Yet when many big-budget AAA games boil down to that samey open-world format, or developers experience huge layoffs weeks after releasing a new title, it can all leave one feeling mightily cynical at the state of things.