AI Assembled 030226

Whether it's Slots or Live Dealers, SpinQuest.com has the fun and action you're looking for,

with SpinQuest exclusives, Blackjack, Roulette, Fakera, and even Live Dice with Craps and

Bubble Craps.

The games never stop, so you don't have to.

And right now, new users get $30 coin packs for just $10 bucks.

Play now at SpinQuest.com.

SpinQuest is a free to play social casino.

Boydware prohibited, visit SpinQuest.com for more details.

Spring times the perfect time to turn the page and refresh your home style.

The only problem, it can be expensive.

Luckily, there's a better way to shop well-worthy pieces for less, at Bob's discount furniture.

Bob's negotiates with manufacturers to get you the best everyday low prices, giving

the shopping power back to you.

The power to get more style, comfort, and quality for less on amazing pieces.

Like stylish mid-century dining sets made for hosting those big spring branches, pop

up sleeper sectionals with all the bells and whistles so you can turn any room into a

guest room and next level Bobo-Pedic mattresses so you can get a great night's rest after

you finish redecorating every room.

So, when you're ready to spring into new style, stop into your nearest store or shop

online and see how you can get well-worthy style for less with Bob's discount furniture.

Welcome to AI Assembled, the top AI news for small business.

I'm Chris Daly, and each week I bring you the most important AI developments that can

transform your business.

Let's dive into this week's top stories.

The first week of March 2026 arrived, carrying the kind of news that arrives only once in

a generation, and then kept arriving.

If you work with AI tools, depend on AI platforms, or simply run a business in a world where

artificial intelligence is becoming impossible to ignore, what happened between February 23rd

and March 2nd belongs on your radar.

Not the highlight reel, the full picture.

Let's start with the model that launched the week's conversation.

Anthropic released Claude Opus 4.6 on February 5th, and by late February, it was clear this

was not a routine update.

The headline feature is a 1 million token context window, five times the previous limit,

which means a single conversation can now hold the equivalent of a full length novel,

an entire code base, or months of business documents without losing track.

On the benchmark that matters most to enterprise users, GDP Val AA, which simulates complex

economic task completion, Opus 4.6 beat GPT 5.2 by 144 elo points.

It topped Terminal Bench 2.0 for coding.

It scored highest on humanity's last exam.

The pricing held, five dollars per million input tokens, 25 per million output.

This is the model and Anthropic is betting its commercial future on, and if the benchmarks

hold in production, that bet has a real chance of paying off.

But the model itself is only part of the story.

Claude Code and Thropic's developer tool built on top of these models shipped six significant

updates in February that change how the tool actually works in practice.

Agent Teams is now available in Research Preview, letting multiple Claude instances collaborate

in parallel on complex coding tasks.

Think of it as assigning a problem to a team rather than a single developer.

One instance handles the architecture, another writes the tests, a third reviews the output.

Automatic memory launches a background process that learns your preferences and your code

base over time, without you having to do anything.

It observes the way you work and builds a persistent context that carries forward across sessions.

Skills auto loading means any skills you drop into a directory are discovered automatically,

which matters if you're running a shared repository across a team.

There are also Cleaner CLI Authentication commands, a native Windows ARM 64 binary for developers

on ARM hardware, session auto naming with slash rename, selective conversation summarisation

and tighter VS code integration.

This is Claude Code moving from a single session coding helper into something closer to a persistent

development teammate.

Running alongside that story is the OpenClaw 2026.2.21 release, which dropped on February 21

and was substantial.

The most significant headline is Gemini 3.1 integration.

OpenClaw now natively supports Google's latest flagship model alongside its existing

Claude and GPT support, expanding the options available to operators, building autonomous

AI workflows.

Security hardening was extensive.

This single release included over 100 fixes targeting known vulnerability patterns, following

a period of escalating scrutiny from researchers and regulators worldwide.

The discord upgrades are worth noting for teams that use it as a coordination layer.

Streaming previews now show inline results, voice channel support landed via a slash VC

command, and notification spam was significantly reduced.

Threadbound sub agent sessions mean that when OpenClaw spawns sub agents, they're now

scoped to individual threads rather than bleeding across conversations, which was a critical

architectural gap.

Which brings us to a related story that has been growing quietly, but carries real consequences.

The rise of managed hosting services for OpenClaw.

Companies like OpenClaw.ai launched this month, offering hosted instances of OpenClaw.

To pay a monthly fee, they handle the deployment and the security patching.

On one level, this is practical.

Not every business has the technical staff to self-host a complex AI orchestration system

and keep it patched against a rapidly evolving threat landscape.

On another level, it directly undercuts what makes OpenClaw valuable.

The open source, self-hosted model that keeps your data, your prompts, and your workflows

off someone else's servers.

Paying a third party to run OpenClaw for you means your most sensitive automation logic

is now in their infrastructure.

The convenience is real.

So is the trade-off.

Every business using a managed OpenClaw service needs to ask exactly the same questions they'd

ask before handing their email to a vendor.

Who has access?

What do they log?

And what happens to your data if the service goes dark?

Meanwhile, the economics of running AI at all continued their dramatic shift.

In Vidya H100 GPUs have lost approximately 85% of their value since peak pricing, from

around $40,000 new to roughly $6,000 on the secondary market.

Cloud rental rates followed the same trajectory.

From $8 per hour in late 2024 to $285 to $350 per hour today, a 64% collapse.

AWS cut its H100 rates by 44% in June 2025.

GCP spot pricing sits around $225.

Over 300 new cloud providers entered the market in 2025, driving competition that is still

compressing margins across the industry.

For small businesses, this means the compute that was prohibitively expensive 18 months

ago is now genuinely accessible.

The political dimension of AI came to a head this week in a way that will matter for years.

The Trump administration, through executive action, effectively blacklisted anthropic

from federal defense contracts.

The stated reason involved compliance concerns, specifically anthropic safety first posture,

which some in the administration viewed as an obstacle to certain military applications.

Within 24 hours of that announcement, open AI had signed its own deal with the Pentagon.

The contrast was not subtle, and the message to the broader AI industry was unmistakable.

Many commitments have a price in the current political environment.

Open AI's contract with the Department of Defense includes what the company described

as layered protections, three explicit red lines prohibiting certain autonomous weapons

applications, with oversight mechanisms built into the agreement.

Whether those protections are substantive or performative is a debate the defense community

is still having.

It is not debatable is the signal.

The federal government is actively sorting AI companies into preferred and non-preferred

vendors.

That sorting will shape federal procurement, enterprise software adoption, and ultimately

which platforms define the standard for AI at scale.

For anthropic, the practical response was immediate.

On February 24th, the company launched an expanded commercial push through its co-work

platform, adding enterprise plugins for finance, HR, engineering and design, alongside

a formal partnership with PWC for finance and life sciences deployment.

When your government revenue is at risk, you move harder on your commercial revenue.

The enterprise market became anthropics insurance policy this week.

Three days before the Pentagon deal was signed, a story broke that reframed every conversation

about AI capability.

Bloomberg reported on February 25th that a hacker had used anthropics Claude code, the

same tool that shipped six new features this month, to steal approximately 150 gigabytes

of data from Mexican government agencies.

The haul included roughly 195 million taxpayer records.

The attacker used Claude's ability to automate complex technical workflows, directing it

to navigate systems and exfiltrate data at a scale that would have required a skilled

team working for weeks.

It happened in a fraction of that time.

This is not a story about Claude code being broken.

It is a story about what happens when any powerful capability, the same one that makes

developers 10 times more productive, lands in hands that want to use it as a weapon.

The same context window and a genetic planning that make it a legitimate productivity tool

also made it a force multiplier for an attacker.

IBM's ex-force report released the same week, documented a 44% year-over-year increase

in AI-assisted attacks.

The pattern is clear, the security conversation cannot be an afterthought.

According to this charged landscape, Gensbark launched an AI browser that frames tool independence

differently.

Available for Windows and Mac, it runs up to 169 open-source AI models entirely on your

device.

No API subscription required.

No data leaving your machine.

No exposure to any of the geopolitical and legal turbulence swirling around the

major platforms.

For businesses watching the open AI Pentagon deal and the anthropic blacklisting and wondering

whether their tools could be caught in similar crossfire, a locally run model browser is a

genuine hedge.

It won't outperform GPT-5 on raw benchmarks, but it offers a kind of sovereignty that

cloud-dependent tools simply cannot match.

Meta Manus, arriving in ads manager this week, represents the other end of that spectrum.

Deep embedding rather than independence.

Meta integrated its Manus AI agent directly into its advertising platform, giving small

business advertisers AI-driven campaign strategy, audience targeting, and optimization inside

the tool where they already work.

No new subscription, no learning curve.

It is genuinely useful.

The question that should accompany that usefulness is, when the AI advising your ad strategy

is built and operated by the platform collecting your ad spend, whose interests is it optimizing

for?

And finally, the Amazon Open AI $50 billion AWS partnership framed the week's financial

architecture.

This is the largest strategic AI partnership in history, and it signals a clear direction.

AI is no longer a separate product category.

It is becoming infrastructure, embedded inside the platforms you already depend on.

That makes it more accessible and harder to walk away from at the same time.

For every small business navigating this moment, the practical work is the same.

Know which platforms hold your data.

Know what you're getting from each AI tool.

And know where you want to keep your options open.

The company's making $100 billion decisions have made their choices.

So what does all of this add up to, practically, for a small business trying to decide what

to do next?

Start with Claude Opus 4.6 and Claude code.

If you have developers on your team, or if you're using any AI coding assistant at all,

this is the moment to test whether the 1 million token context window changes, what you can

actually accomplish.

The ability to hold an entire codebase in context changes how you debug, how you refactor,

how you scope new features.

That is not a marginal improvement.

It is a structural one.

Try it on your most complex project, not your simplest.

On OpenClaw, if you are running a self-hosted installation, the 2026.221 update is not optional.

The security hardening addresses vulnerabilities that researchers have already published.

If you are considering a managed hosting service instead, treat it like any other vendor that

would have access to your operational data, which means a security review, a data handling

agreement, and an honest assessment of whether the convenience is worth the exposure.

The GPU price story has a direct implication that is easy to miss.

If you have been deferring AI experiments because the compute costs seemed prohibitive,

those costs are now a fraction of what they were.

A budget that would have rented 10 hours of H100 time in 2024, now by 60 or more.

This is the window to run experiments that your organisation previously couldn't afford.

The political story, Anthropics Blacklisting, OpenAI's Pentagon deal, the Amazon AWS

consolidation, is slower moving, but worth watching.

The decisions being made right now about which AI companies are federally preferred are

going to shape enterprise procurement patterns for the next several years.

If you have federal customers or government contracts, understanding your AI vendor's

political standing is now part of your vendor management process.

And finally, the security picture from Mexico and the IBM X-Force report has a simple

take away.

The AI tools that make your team more capable are the same tools being used against systems

like yours.

The entry point in most attacks isn't a sophisticated exploit, it's an unpatched application, a reused

password, or a staff member who didn't recognise a convincing AI-generated phishing message.

The technical sophistication of the attack has increased dramatically.

The defensive basics have not changed.

That is the AI landscape as of March 2nd, 2026.

It is more powerful, more contested, more affordable, and more dangerous than it was this time

last year.

Navigate it with both eyes open.

That's it for this week's AI Assembled, the top AI news for small business.

If you found this valuable, share it with a fellow business owner.

If you want a hands-on group that you can work with on your specific AI needs, consider

joining Business Success Academy at freshmediaworks.com slash BSA.

Until next week, keep innovating and keep it real.

Forget whatever plans you have this weekend because you're staying at home and playing

on SpinQuest.

And there's never been a better time to sign up in right now.

New users get $30 coin packs for just $10, all the table games you love, with hundreds

of slot games and real cash prizes.

That's at SpinQuest.com, SPINQST.com.

SpinQuest is a free-to-play social casino, voidware prohibited, visit SpinQuest.com for more

details.

Guys, it's no use putting it off.

The best time for an underwear refreshes now.

Tommy John underwear is designed for a perfect fit that stays put all day.

There's zero shape thanks to four times more stretch than competing brands, and their

innovative horizontal quick draw fly is a game changer.

With over 30 million pair sold, there are thousands of men out there more comfortable

than you.

Don't settle for less.

Go to TommyJohn.com today for 25% off your first order with Code Comfort.

That's TommyJohn.com Code Comfort.

Tommy John.

Comfort.

Perfected.