OpenAI launched ChatGPT on November 30, 2022. We started using it nine days later, on December 9, 2022. Since then, we have run AI tools through real client work every single day. That gives us a useful vantage point on how fast the technology has actually moved, and on what is worth paying attention to right now versus what is hype.

Most of the public conversation about AI has stayed at the surface. Faster models. New chatbots. Better answers. Those are real, but they are not the headline. The bigger story in 2026 is that AI has crossed several thresholds that change how the technology fits into everyday work and everyday risk. Three of those thresholds matter most for small businesses in Fargo-Moorhead right now.

AI Video Is Now Hard to Tell From Real Footage

Two years ago, AI-generated video looked like a glitch. The famous “Will Smith eating spaghetti” clip from 2023 became a meme because the proportions were wrong, the motion was stuttered, and the food behaved like clay. Anyone could spot it in a second.

That gap has closed faster than almost any other capability in AI. Sora 2, which OpenAI released in September 2025, generates 4K video up to 25 seconds long with realistic physics, light, and motion. Google’s Veo 3.1 followed in January 2026 with native vertical video and continuous narrative scenes longer than a minute. Runway’s Gen-4.5, released in December 2025, added in-video editing where you can change one element with a text prompt without regenerating the whole clip.

The detection numbers tell the rest of the story. A 2026 University of Florida study found that automated tools spot fake faces in still images with 97 percent accuracy, which is now better than the average person. University at Buffalo researchers concluded that in everyday scenarios like low-resolution video calls and social media clips, synthetic video has become indistinguishable from real footage for most viewers, and in some cases even for institutions trying to verify it.

What does this mean for a small business? Two practical things. First, a video on its own is no longer proof of anything. If you receive a video of someone you trust asking for money or credentials, that is not enough to act on. Second, video content is now a viable production channel for small teams. A polished product demo or training clip that used to require a film crew can now be assembled in an afternoon.

AI Has Flipped From a Security Risk to a Security Advantage

For a while, AI was mostly a problem for security teams. It wrote convincing phishing emails. It generated malware variants faster than analysts could review them. It helped attackers find soft targets. Microsoft reported that 41 percent of zero-day vulnerabilities discovered in 2025 were found by attackers using AI tools, before defenders had a chance to see them.

That equation flipped in April 2026. Anthropic announced Claude Mythos, a research model focused on security. In controlled testing, Mythos identified thousands of previously unknown vulnerabilities across major operating systems and web browsers, including a 27-year-old flaw in OpenBSD, a 16-year-old issue in the widely used FFmpeg media library, and a 17-year-old remote code execution bug in FreeBSD. Each had survived nearly three decades of human review.

Rather than release Mythos broadly, Anthropic created Project Glasswing, a consortium of more than 40 organizations including Apple, Microsoft, Google, and AWS. Members get monitored access to use the tool defensively, patching their systems before similar capabilities reach the attacker side. The bet is straightforward: defenders should get these tools first.

Gartner projects that more than 60 percent of organizations will use AI-augmented security platforms by the end of 2026, up from under 20 percent in 2023. The practical takeaway for a small business is that AI is now the strongest defensive layer you can put in place. Modern endpoint protection, email filtering, and patch management tools that use AI find threats faster and miss fewer of them than the rule-based tools they replaced.

AI Can Now Build Working Software, Not Just Snippets

In 2023, AI could help you write a function. By late 2024, it could draft small features. In 2026, frontier models can take a written description and produce a working application, complete with tests and deployment configuration, often with no manual code edits required at all.

Claude Opus 4.7, released by Anthropic in April 2026, scores 87.6 percent on SWE-bench Verified, a benchmark that measures whether a model can correctly fix real bugs in real open source projects. A year ago, the leading scores on the same benchmark were in the 30s. Mythos, the research model from the same announcement window, pushed past 93 percent.

The real-world results match the benchmarks. Rakuten reported that feature delivery time on one of its products dropped from 24 working days to 5 working days after the team built Claude Code into their workflow, a 79 percent reduction. Boris Cherny, who leads the Claude Code product at Anthropic, has written 100 percent of his code contributions through Claude Code since November 2025, with zero manual edits.

For a small business, this changes the economics of custom software. A custom inventory tracker, a client portal, an internal reporting dashboard, or a process automation script that would have cost tens of thousands of dollars three years ago is now within reach for a fraction of that. The work still needs someone who can write a clear specification and verify the output, but the build step itself has collapsed.

Fargo-Moorhead Is Part of the Story

These breakthroughs are not happening somewhere far away. Applied Digital is building the Polaris Forge 2 data center in Harwood, North Dakota, just north of Fargo. The facility will carry 200 megawatts of computing capacity dedicated to AI workloads, run on a $5 billion 15-year contract with a major hyperscaler, and create more than 200 permanent jobs once operational. Initial capacity comes online late 2026.

North Dakota State University is responding on the education side. Beginning fall 2026, NDSU will award full-ride scholarships to 30 incoming students per class for a new Honors College focused on AI ethics, technology, and society. The university is also launching a dedicated AI major and recently received funding to build a research supercomputer named Bison. The combined effect is that the region is producing the infrastructure and the people to use it at roughly the same time.

For local businesses, that means the gap between hearing about an AI capability and being able to apply it in Fargo-Moorhead is closing fast. The technical support, the talent pool, and the regional momentum are all showing up together.

What This Means for a Small Business Right Now

None of the breakthroughs above require an in-house data science team. They show up in tools you already use or can buy. Better video creation lives inside Canva, Adobe, and Runway. AI-driven security is built into modern Microsoft 365, Cloudflare, and managed IT offerings. AI-assisted coding shows up in custom-built business tools that used to be off-limits to small operations.

We offer classes that walk through how these tools fit into real Fargo-Moorhead businesses. The classes focus on practical use rather than theory, and on the small set of skills that actually matter: writing clear instructions, verifying the output, and knowing when to trust the result. The learning curve is much shorter than most people expect.

The harder question is the one only you can answer for your business. Which task is taking too long? Which risk has been on your list for months without a good answer? Which capability would change the way your team competes if you had it tomorrow? Those answers point straight at where AI is most useful right now, in 2026, with the tools that already exist.

The breakthroughs have happened. The next step is using them.