Author: Shahzad Khan

When Sam Altman predicted that AI would enable a one-person billion-dollar company, many assumed it would come from breakthrough technology or a revolutionary product.

Instead, the first real example looks very different.

From $20K Experiment to $1.8B Trajectory

According to reporting from The New York Times, entrepreneur Matthew Gallagher scaled his startup Medvi from a $20,000 experiment into a business projected to reach $1.8 billion in annual sales.

Even more striking:

• The company generated $401 million in revenue in its first year
• The initial build took just two months
• The core team started as essentially one person

This is not a traditional startup story. There was no large engineering team, no years of R&D, and no massive VC-backed runway.

The Business Model: AI + Execution

Medvi operates in the telehealth space, selling GLP-1 weight-loss medications online.

Instead of building everything from scratch, Gallagher leveraged existing platforms:

• Telehealth providers like CareValidate and OpenLoop handled doctors, prescriptions, and compliance
• Logistics and fulfillment were outsourced
• The business focused on distribution, marketing, and orchestration

This is a key shift: AI didn’t replace the system — it orchestrated it.

The AI Stack Behind the Growth

Gallagher used a combination of AI tools to replace what would traditionally require entire departments:

• ChatGPT, Claude, and Grok for coding and automation
• Midjourney and Runway for ad creatives
• ElevenLabs for customer interaction
• Custom AI agents for support, workflows, and operations

The result: a lean, AI-augmented operation with minimal human overhead.

Team Size: Almost Non-Existent

After scaling, Gallagher added just one full-time employee—his brother.

Everything else runs through:

• Contractors
• External partners
• AI systems

This is a radical departure from the traditional startup scaling model, where headcount grows alongside revenue.

Why This Matters

This story challenges a common assumption: that massive outcomes require massive teams.

Instead, it highlights a new model:

AI + distribution + execution > large teams + long timelines

And perhaps the most surprising part:

This isn’t a deep-tech AI company.
It’s a distribution business powered by AI tools.

The Real Insight

The breakthrough isn’t the product—it’s the operating model.

• AI compresses time (2 months to launch)
• AI reduces cost ($20K to start)
• AI replaces roles (engineering, marketing, support)
• Platforms handle infrastructure (telehealth, logistics, compliance)

What’s left is decision-making, direction, and execution.

Final Thought

The first AI-powered billion-dollar company didn’t come from a lab.

It came from someone who understood how to combine tools, platforms, and speed.

And that raises an uncomfortable but important question:

How many “impossible” businesses are now just execution problems?

https://www.nytimes.com/2026/04/02/technology/ai-billion-dollar-company-medvi.html

In a striking vision of the future of work, Jack Dorsey—co-founder of Twitter and CEO of Block, Inc.—has put forward a provocative idea: AI can replace middle management.

This isn’t just theory. It’s already being tested in practice.

---

The Shift at Block

Earlier this year, Block reduced its workforce by over 4,000 employees—more than 40% of its staff. According to Dorsey, this wasn’t a reaction to financial distress, but a deliberate move toward an AI-first organization.

The company is restructuring around a leaner, more focused model where traditional management layers are no longer central.

Instead, Block now defines its workforce across three roles:

• Builders – individuals who create products and systems
• Problem Owners – those accountable for outcomes and results
• Player-Coaches – experienced contributors who guide and mentor others

This model removes the need for conventional middle managers whose primary role has historically been coordination.

---

Why Dorsey Thinks AI Can Replace Managers

Dorsey’s argument is rooted in how modern organizations already operate—especially remote-first ones.

At Block, nearly everything is documented digitally:

• Decisions
• Product designs
• Internal discussions
• Strategic plans

This creates a rich dataset—a living “world model” of the business.

According to Dorsey, AI systems can now:

• Track and interpret this information in real time
• Route insights and updates across teams
• Identify bottlenecks and inefficiencies
• Provide decision support at scale

In short, AI can perform one of the core functions of middle management: information flow and coordination—but faster and without organizational friction.

---

The Bigger Picture: A New Organizational Model

Dorsey’s thesis reflects a broader trend:

Lean, AI-first teams vs. traditional, layered enterprises

In traditional organizations:

• Information moves slowly through hierarchies
• Decision-making is fragmented
• Accountability is often diffused

In AI-enabled organizations:

• Data is centralized and continuously analyzed
• Decisions can be made closer to the work
• Teams operate with greater autonomy

This could lead to:

• Faster execution
• Lower operational overhead
• More direct ownership of outcomes

---

The Risks and Open Questions

Despite the promise, this shift raises important concerns:

1. Trust in AI Decision-Making

Organizations may hesitate to rely fully on AI for coordination and judgment—especially in high-stakes environments.

2. Loss of Human Context

Middle managers often provide:

• Emotional intelligence
• Conflict resolution
• Cultural alignment

These are areas where AI still has limitations.

3. Organizational Stability

Flattening structures too aggressively can lead to:

• Role ambiguity
• Burnout among high performers
• Gaps in leadership development

---

What This Means for Professionals

If Dorsey’s model gains traction, the implications are significant:

• Execution > Coordination: Value shifts toward building and delivering
• Ownership becomes critical: Individuals are accountable for outcomes, not just tasks
• AI fluency is no longer optional: Understanding how to work alongside AI becomes a core skill

For engineers, architects, and platform-focused professionals, this may actually be an advantage—especially those already working close to systems, automation, and outcomes.

---

Final Thought

Dorsey’s vision challenges a long-standing assumption: that organizations need layers of management to function effectively.

Instead, he proposes a future where:

AI becomes the connective tissue of the company, and humans focus on creation, ownership, and growth.

Whether this model scales across industries remains to be seen. But one thing is clear—the role of middle management is being fundamentally re-evaluated in the age of AI.

https://block.xyz/inside/from-hierarchy-to-intelligence

OpenAI just raised $122B at an $852B valuation — the largest venture funding round in history — and it’s signaling something much bigger than just capital.

This is a strategic shift toward an “AI superapp.”

What stands out:

• Amazon, Nvidia, and SoftBank anchored ~$110B of the round
• Amazon reportedly included an AGI-trigger clause — a fascinating signal about where this is headed
• Revenue has reached $2B/month, growing at a pace 4x faster than early-stage Alphabet and Meta
• Enterprise now drives 40%+ of revenue, on track to match consumer soon
• ChatGPT, Codex, and agent tools are being merged into a single unified platform
• Meanwhile, side efforts like Sora are being deprioritized

Why this matters:

This isn’t just about scale — it’s about focus.

The real story is the enterprise shift. When nearly half of revenue is coming from enterprise (and rising), it tells you exactly where the durable value is being built.

We’re watching the transition from:
➡️ AI tools →
➡️ AI platforms →
➡️ AI operating systems for work

The “superapp” direction suggests a future where:

• Development, automation, and decisioning live in one interface
• Agents become first-class coworkers
• AI moves from assistive → operational

If this trajectory holds, the next phase of the AI race won’t be about who has the best model — it will be about who owns the workflow layer.

And that’s where the real competition begins.

https://openai.com/index/accelerating-the-next-phase-ai

Details surrounding Anthropic’s next flagship AI model—reportedly named Claude Mythos—have surfaced following an apparent internal misconfiguration that exposed unpublished launch materials.

According to the leaked draft blog and supporting assets, Mythos is positioned as a significant advancement over the current Claude lineup, described internally as “a step change” and potentially the company’s most capable system to date.

What Happened

The exposure appears to stem from a CMS configuration error that left thousands of internal assets accessible through a public data cache. Among them was a draft announcement detailing Mythos and its capabilities.

While such leaks are not unheard of in the AI industry, the nature of the content—particularly around safety and cybersecurity—has drawn notable attention.

A New Tier Above Opus

One of the most striking revelations is the introduction of a new model classification tier, internally referred to as “Capybara.”

This tier is said to sit above Anthropic’s existing Opus class, implying:

• Larger and more complex model architecture
• Higher computational cost
• Expanded capabilities across reasoning and coding

If accurate, this signals a continued vertical scaling strategy among frontier AI labs, where each generation pushes beyond prior limits in both performance and resource intensity.

Cybersecurity Capabilities Raise Concerns

The leaked materials reportedly highlight Mythos as being “far ahead of any other AI model in cyber capabilities.”

This includes the potential to:

• Identify vulnerabilities more effectively
• Assist in advanced exploit development
• Accelerate offensive security workflows

Anthropic’s internal language also acknowledges the dual-use risk—warning that such capabilities could enable attackers to outpace defenders if not carefully controlled.

Official Confirmation (Without the Name)

In response to inquiries, Anthropic confirmed to Fortune that it is actively testing:

“a new general purpose model with meaningful advances in reasoning, coding, and cybersecurity.”

Notably, the company did not confirm the Mythos name or the leaked tier structure, but the description aligns closely with the exposed materials.

Why This Matters

This incident highlights several important trends in the AI landscape:

1. The Frontier Is Still Accelerating
A new tier beyond Opus suggests that major labs are continuing to push the boundaries of scale and capability, not slowing down.

2. Cybersecurity Is Becoming a Core AI Battleground
Models are no longer just productivity tools—they are increasingly capable of participating in both defensive and offensive security workflows.

3. Safety vs. Capability Tension Is Growing
For a safety-focused organization like Anthropic, the leak raises questions about how such powerful systems are controlled, tested, and eventually released.

4. Strategic “Leaks” and Industry Hype
Whether accidental or not, the situation echoes past incidents—such as OpenAI’s Q*-era rumors—where early disclosures amplified anticipation and shaped industry narratives.

Final Thoughts

If Claude Mythos—or whatever the final release is called—delivers on the leaked claims, it could represent another major inflection point in AI capability.

But with that leap comes increased responsibility.

The real question is no longer whether AI systems can reach these levels of capability—it’s how the industry will manage the risks that come with them.

Claude Mythos Blog Post

Saved before it was taken down.https://t.co/6XIw1LKnkA

— M1 (@M1Astra) March 27, 2026

4

Meta has taken a bold step into the future of neuroscience with the release of TRIBE v2—an open-source AI model that can simulate human brain activity across vision, hearing, and language. What makes this breakthrough remarkable isn’t just its scale, but its performance: in some cases, its synthetic predictions outperform actual fMRI brain scans.

This signals a potential turning point where software begins to rival—and even replace—traditional brain imaging experiments.

---

🚀 What TRIBE v2 Actually Does

TRIBE v2 is designed to model how the brain responds to different stimuli—like images, sounds, and text—without needing a human subject inside an MRI machine.

Here’s what sets it apart:

• Massive scale-up in data and scope
  • Trained on 1,000+ hours of brain recordings
  • Expanded from 1,000 → 70,000 brain regions
  • Built using data from 700+ individuals (vs. just 4 in v1)
• Cross-modal intelligence
  • Simulates neural responses across:
    • 👁️ Vision
    • 👂 Hearing
    • 🗣️ Language
• High-fidelity predictions
  • Its outputs align with population-level brain activity
  • In some cases, cleaner than real fMRI scans, which are often noisy due to:
    • Heartbeats
    • Movement
    • Scanner artifacts

---

🧪 A Surprising Result: AI vs. Real Brain Scans

One of the most striking findings is that TRIBE v2 can outperform actual fMRI data in predicting brain activity patterns.

That sounds counterintuitive—until you consider:

• fMRI scans are inherently noisy and indirect
• AI models can produce clean, idealized signals
• Aggregated training across hundreds of people removes individual variability

In effect, TRIBE v2 creates a “denoised, generalized brain”—something neuroscientists have never had access to before.

---

🧠 Reproducing Decades of Neuroscience—Without Scans

Perhaps the most impressive capability: TRIBE v2 can rediscover known brain mappings purely in software.

Without running new scans, it correctly identified:

• Face-processing regions
• Speech-related areas
• Text and language centers

This means the model has internalized fundamental principles of brain organization—a milestone for computational neuroscience.

---

🔓 Fully Open-Source (and That’s a Big Deal)

Meta didn’t just publish a paper—they released:

• ✅ Model weights
• ✅ Source code
• ✅ Live demo environment

This dramatically lowers the barrier to entry. Researchers no longer need:

• Access to expensive MRI machines
• Complex experimental setups
• Large subject pools

Instead, they can run virtual brain experiments on demand.

---

⚡ Why This Matters (The AlphaFold Moment?)

This could be neuroscience’s version of AlphaFold.

Before AlphaFold:

• Protein research required years of lab work

After AlphaFold:

• Structures can be predicted in minutes

TRIBE v2 could trigger a similar shift:

Traditional Neuroscience	With TRIBE v2
Expensive MRI scans	Virtual simulations
Weeks/months per study	Seconds/minutes
Limited sample sizes	Scalable datasets
High noise levels	Clean predictions

---

⚠️ Important Caveats

Despite the excitement, this isn’t a full replacement for real neuroscience (yet):

• It models average brain behavior, not individual differences
• It depends heavily on training data quality
• Real-world validation is still essential

Think of it as a powerful accelerator, not a total substitute.

---

🧭 The Bigger Picture

TRIBE v2 hints at a future where:

• Brain research becomes compute-driven instead of hardware-limited
• Hypotheses can be tested before involving human subjects
• AI helps uncover patterns we might never detect manually

For someone like you—working in Azure + AI systems design—this is also a signal:

👉 The next wave of AI isn’t just language or vision—it’s biological system simulation at scale.

---

💡 Bottom Line

TRIBE v2 is more than a model—it’s a shift in how we approach understanding the brain.

If it continues to evolve, we may soon reach a point where:

• Running a neuroscience experiment
• Feels more like running a cloud workload

And that’s a profound change.

https://ai.meta.com/research/publications/a-foundation-model-of-vision-audition-and-language-for-in-silico-neuroscience