Michael Janis

Exploring the future of value in the AI era
Investor's mindset, builder's skillset

[Michael Janis @michaeljjanis]

@TiffanyFong_ Vices are always profitable

[Michael Janis @michaeljjanis]

The Era of Compounding Software

Enterprise software suffers from a fatal structural flaw: forced consensus. To achieve commercial scale, enterprise developers must average out the workflows, proprietary mental models, and edge-case brilliance of ten thousand different users. The result is a rigid assembly line. SaaS companies are struggling today because AI is actively commoditizing these generic workflows.

We are entering the era of Compounding Software—hyper-personalized, adaptive architecture that acts as a literal cognitive sponge for specialized domain expertise.

Historically, creating software that adapted to the individual user simply did not scale. A few elite technology firms managed to build bespoke systems for massive institutions, but they achieved it through brute force. They deployed foundational data engines and then sent armies of highly paid, forward-deployed human engineers to manually customize the codebase to the client's unique ontology. It was a high-cost service model, impossible to deploy to the masses.

This exact model can now be executed autonomously at the individual user level. The platform acts as the core engine. An embedded AI serves as the dedicated forward-deployed engineer, continuously rewriting the codebase in real time to fit the operator.

Mapping messy, real-world complexity requires crossing a strict engineering limit to make this safe and functional. Deterministic architectures, like knowledge graphs, are perfectly grounded. They return statistically true intersections, delivering raw signal. Generative AI possesses deep reasoning and human-level context.

What we have learned in building these systems is that math and context are entirely codependent. Math without context is just noise. It calculates the physical coordinates of a connection but provides no understanding of why it matters. Context without math is a hallucination. It provides a brilliant narrative with no structural reality.

However, when properly architected, hallucination transforms into a massive asset. You can allow the Generative AI to dream—to generate wild, non-intuitive, high-variance hypotheses. You then immediately ground-truth those outputs against the deterministic math model. Hallucination functions as a highly efficient engine for novel discovery, provided the math is there to ruthlessly score and validate the output.

The structural alpha lies in the absolute fusion of the two. Deterministic math builds the rigid scaffolding—the absolute walls of reality. Generative AI then operates strictly inside that boundary to reason, filter, and extract meaning. This hybrid architecture grounds the system and allows the software to safely adapt.

The adaptation happens through captured data exhaust. As a domain expert simply does their job—flagging a false risk in a contract, injecting a supplier preference, or validating a non-standard valuation metric—the embedded AI agent observes the behavior. The system autonomously translates those daily human interactions into recursive updates to its own logic. The software actively learns both the structural realities of the domain and the procedural workflow of the human operating it.

Deploy the exact same baseline software to a thousand domain experts. Six months later, the result is a thousand distinctly evolved architectures. The software biologically diverges to match the unique intellectual ontology of its operator. Two competing organizations processing the exact same public data streams will extract completely different, highly specialized conclusions.

The ultimate enterprise moat is the continuous institutionalization of human expertise. Capturing the data exhaust from every domain expert creates a compounding digital asset. Your software must get smarter every time your best people touch it.

When this happens, the fundamental calculus of corporate competition changes. If a rival attempts to catch you, they cannot simply buy the same off-the-shelf SaaS license. They are mathematically locked out, years behind in the evolutionary cycle of the software itself.

The most valuable companies of the next decade will not be invested in or acquired for their physical infrastructure, their market share, or even their proprietary data lakes. Buyers will acquire companies simply to extract the compounding digital brain the organization accidentally grew while its workforce was busy doing their jobs.

[Michael Janis @michaeljjanis]

@elonmusk I showed up to vote in Orange county, took out my id, and was told that they weren't allowed to look at it

[HHS Rapid Response @HHSResponse]

The problem? 90% of healthcare spending goes to chronic disease. The solution? EAT REAL FOOD. 🎥

[Michael Janis @michaeljjanis]

@ClaudiaAna98888 Excellent point

[Michael Janis @michaeljjanis]

The Rise of the Alien Competitor

I believe we are staring at a Speciation Event in the corporate world. A chasm is opening between the "Old Guard" and a new breed of competitor. And this new breed isn't just faster than us.

In fact, to any traditional organization, they will look entirely Alien.

Constraint Thinking

To understand why, look at how corporate operators are trained. Traditional engineering and business programs train us to manage limitations. We obsess over memory allocation, code efficiency, and budget optimization.

We are trained to find the "best" solution given the strict constraints of the tool.

Whether it is the row limit of a spreadsheet, the compute power of a server, or the physical throughput of a global supply chain, we are conditioned to shrink the problem until it fits the box we have.

But what happens when the box disappears? What happens when intelligence is no longer a monolithic constraint, but an infinite, disposable resource?

An "Alien" company is one that stops solving for Scarcity (how do I fit this problem into my tool?) and starts solving for Abundance (what is the answer if the tool has no limits?).

Two Alien Thought Experiments

1. Scientific & Material R&D

The Traditional Approach: A team of brilliant researchers brainstorms a difficult scientific or materials target, filtering down to three plausible candidates based on intuition and past heuristics. They optimize for Probability—picking the horse most likely to win because the R&D budget only allows them to run one race.

The Alien Approach: You spin up 100,000 AI agents—each with PhD-level domain reasoning—for the cost of an executive lunch. You do not ask them to be "right"; you ask them to be exhaustive. You generate 5,000 counter-intuitive materials or compounds that look "wrong" to human bias. You simulate them all. You find the one bizarre outlier that breaks the paradigm. The Alien optimizes for Variance.

2. Physical Hardware & Engineering

The Traditional Approach (Iterative Design): A legacy manufacturer designs a new aerospace bracket or industrial heat exchanger. They rely on symmetry and decades of heuristics. They tweak the geometry in CAD, optimizing for Feasibility—staying strictly within the bounds of what they know can be machined or cast.

The Alien Approach (Inverse Design): Rather than designing the part, you define the Outcome—the exact thermal dissipation or structural load capacity required. You unleash an AI to evolve the geometry required to create that physics state. It spits out an asymmetric, organic, non-intuitive shape that looks "broken" to the human eye, but it stabilizes the physics in a way standard engineering never could. The Alien optimizes for Absolute Performance.

The Physics of Profit

There is a financial reality to this speciation event that we have to appreciate.

Gross Margin is base value creation. It represents the raw utility extracted from new products and services.

OpEx is optimization. It represents the cost of the corporate metabolism.

Mutation (the Alien approach) attacks Gross Margin. It discovers new structural geometries, breakthrough therapeutics, and entirely new markets. It fundamentally extends the surface area of value creation.

Efficiency attacks OpEx. It squeezes the metabolism to preserve the value already created.

We cannot cut OpEx into a new future. Optimization has a mathematical limit (zero cost). Value creation has no limit. The Alien company wins because they use Intelligence to radically expand the surface area of their Gross Margin in ways a human workflow never could.

Efficiency subtracts. Mutation multiplies.

There is also a human implication here that we should be honest about.

Efficiency is fundamentally subtractive. It asks: "How can we do the same work with fewer resources?" That helps the bottom line, but it has a limit.

Mutation is fundamentally expansive. It asks: "If intelligence is free, what new branches of physics, material science, and biology can we colonize that we couldn't touch before?"

The "Alien" company doesn't use AI to cut headcount. It uses AI to massively expand its surface area—to run more experiments, enter more markets, and test more "weird" ideas than a human organization could ever manage.

Escaping the Gravity Well

The gravity of the corporate world pulls relentlessly toward Efficiency. It is measurable. It is safe. It fits neatly into a quarterly review.

Companies that over-index on the Efficiency path are choosing a slow fade. They are optimizing themselves into irrelevance against competitors who are playing a different game entirely. The gap will widen—slowly at first, and then more rapidly than can be overcome with any amount of capital, creating an inescapable event horizon.

To escape the gravity well of traditional efficiency, companies must implement the kind of radical variance that allows them to survive the speciation event.

[Michael Janis @michaeljjanis]

@MikeHoffmann I think it's the opposite - AI allowing execution of so many ideas at the 'edge' - where innovation happens

[Michael Janis @michaeljjanis]

Antigravity IDE at home and Antigravity CLI on the road with SSH and seamless access to the same context window. easy to tell it was made by developers for developers - thank you @googledevs !

[Michael Janis @michaeljjanis]

stop what you are doing right now and take your hand and place it atop your head. You are holding everything in your universe, everything you've been and ever will be. More efficient and powerful than anything being developed by any AI lab in Palo Alto. You are special.

[Michael Janis @michaeljjanis]

The Half-Life of Software: Why Innovation Belongs at the Edge

In my last post, I discussed "Data Intentionality." Today, I want to explore a conceptual shift that challenges our fundamental operational assumptions: The Half-Life of Software.

There was a watershed moment recently that should catch the attention of every enterprise leader. CNBC anchor @dee_bosa—who is not a developer—built a functional replacement for a major enterprise SaaS platform in a weekend using generative AI.

This signals a collapse in the unit economics of software. But the implication is deeper than just "coding is cheap." It suggests that the era of the "Application" as a capital asset is over.

From Monuments to Consumables

For the last twenty years, the corporate world has treated software like capital infrastructure. We built "monuments"—massive, centralized management tools such as ERPs and CRMs designed to last a decade. Because they were expensive to build, we had to centralize control, prioritize rigid roadmaps, and amortize the cost over years.

But business—much like biology—doesn't evolve in the center. Evolution happens at the edge. It is messy, redundant, and hyper-reactive to the environment.

In the AI era, software is no longer a monument. It is a consumable. We are entering the era of Disposable Software: a temporary assembly of logic and data, spun up to solve a specific operational problem, and dissolved the moment that problem changes.

The Friction of the Old Model

This explains why the traditional IT model feels so heavy right now. The "Centralize and Prioritize" framework is breaking because it was designed for long-half-life assets, not the realities of real-time market friction.

A centralized IT team, no matter how brilliant, cannot see the ground-level nuance of a supply chain bottleneck in logistics or a pharmacokinetics anomaly in R&D. When a company forces those problems into a central queue, that company is effectively saying that their internal bureaucracy is more important than the market's reality.

As @JeffBezos famously said, "Your margin is my opportunity." Every friction point a company's employees face—every clumsy workaround, every manual spreadsheet—is margin leaking out of that company. If companies don't give the Edge the tools to fix those leaks instantly, they lose the OODA (Observe, Orient, Decide, Act) loop. When multiplied across an enterprise, this latency becomes existential.

It isn't that IT is executing poorly - rather, it's that the historical model was engineered for a completely different asset class - the model was designed for "Long Half-Life" assets. When a company forces a short-half-life problem (a worker needs a quick data parser now) into a long-half-life queue (a 12-month IT roadmap), it creates severe structural friction. We ask our domain experts to wait for a highway to be built when all they need is a path through the woods.

Reframing IT: The Architecture of Trust and the Physics of the Edge

To move fast, companies must recognize that coding ability is no longer the scarce resource—domain knowledge is. The nuance of a specific workflow lives exclusively at the edge.

This requires a fundamental shift in enterprise roles. IT must transition into the ultimate Enabler, carrying the heavy responsibilities of scalability and security, but with an entirely new purpose:

IT builds the "Architecture of Trust": They govern the secure API gateways, the data ontology, and the governance guardrails. They ensure the environment is robust and secure.

The Edge runs the "Experiments": Empowered by those tools, domain experts build the disposable solutions they need to win today. The Edge ensures the actions deliver immediate, localized value. Let's call it REDI: Rapid Empowered Distributed Innovation.

The Call to Courage

There is a natural corporate reflex to lock this down. To gravitate toward standardization, justification, and cost control. Standardization feels safe; chaos feels risky.

But executive teams must get comfortable with a "Messy Edge." You need to empower your financial analyst or lead scientist to spin up a bespoke AI workflow on Tuesday, solve a million-dollar inefficiency, and delete the tool on Friday.

We are moving into a world where competitive advantage comes from how quickly a company can turn "data exhaust" into domain-specific tools. In a world where software is disposable, the company that empowers its edge to build the fastest won't just be more efficient—it will be a different species of competitor entirely.

[Michael Janis @michaeljjanis]

@trikcode well, we all might be...

[Michael Janis @michaeljjanis]

@BenjaminDEKR nostradamus it's great to hear from you

[Michael Janis @michaeljjanis]

The Efficiency Trap: Why "Better, Faster, Cheaper" Isn't Enough for AI

Recently I've been musing on the directions emerging in enterprise AI. Across every sector, two distinct corporate strategies are emerging—and one of them is a trap.

The first is Automation for Efficiency. This is the standard operational play: using AI to optimize existing workflows, reduce manual toil, and protect current margins. It is necessary, tangible, and highly quantifiable.

The second is Augmentation for Innovation. This is the strategic play: using AI to create differentiated capabilities, products, or insights that were previously impossible or computationally out of reach.

This creates a sharp strategic tension between protecting margins and creating net-new value:

-- Automation creates speed. It takes a standard 4-hour technical analysis and completes it in 4 minutes. This is vital corporate hygiene.

-- Innovation creates capability. It doesn't just speed up a process; it alters the fundamental limits of the product itself.

Most executive teams naturally gravitate toward Automation because the short-term ROI is easy to calculate on a spreadsheet. But Automation inevitably commoditizes.

If every player in an industry lowers their OPEX by 15% using the same off-the-shelf LLMs, no one wins a durable competitive advantage. The playing field simply resets at a lower price point. Efficiency tools are accessible to everyone; therefore, they cannot build a moat.

The real "alpha" lies in Innovation. It is harder to quantify and harder to execute, but it is the path that leads to new revenue, new markets, and new business models.

First-Principles Hardware Disruption

To understand this shift, consider a core structural question: Can you use AI software to make cheap, basic physical hardware perform like an incredibly expensive, specialized instrument?

The answer is yes, and it starts with a first-principles design choice.

Look at the autonomous vehicle sector. Legacy automakers assumed that to achieve autonomy, a vehicle required incredibly expensive, specialized hardware—specifically, active LiDAR sensors costing thousands of dollars per car.

Tesla took a radically different architectural path. Their design choice was rooted in a clean-sheet premise: the entire global road infrastructure was built for human eyes (passive optical sensors) and human brains (neural networks). Therefore, instead of scaling physical hardware costs with LiDAR, they chose basic, low-cost optical cameras (cheap hardware) and augmented them with a massive vision-processing neural network to reconstruct a high-fidelity 3D vector space in real time. The software artificially upgrades the capability of the physical hardware.

This concept—the "Virtual Instrument"—is now poised to completely disrupt the most capital-intensive sector in the world: healthcare and clinical diagnostics.

Breaking the Medical Insurance Bottleneck

Right now, clinical diagnostics are fundamentally broken by hardware economics. High-fidelity medical imaging, complex molecular testing, and early-stage disease screenings cost thousands of dollars per test. Why? Because hospital networks must amortize the multi-million dollar cost of the physical machinery.

Because the upfront cost is astronomical, insurance providers aggressively deny coverage for advanced screenings, leaving patients locked out of early, actionable data until a disease has already progressed.

The next great medical moats will not be built by companies manufacturing heavier, more expensive diagnostic hardware. The market is waiting for AI-augmented instruments.

By leveraging lower-cost, highly accessible physical hardware (like portable, low-Tesla magnets or basic digital assays) and layering them with deep epistemic reasoning networks, we can mathematically reconstruct clinical-grade, high-fidelity results. The intelligence shifts from the physical machine to the digital network.

The Problem of "Data Exhaust"

The bottleneck to executing this strategy across any industry is how companies view information. Historically, physical product, medical device, and industrial companies have treated data as "exhaust"—it is merely the waste byproduct that comes out of a machine while it operates. Companies sell the physical box; the customer walks away with the data.

An algorithm is rapidly becoming a commodity. The real enterprise value is in the ontology—the intentional capturing, structuring, connecting, and mapping of data to reveal insights that were previously invisible.

To build a true AI moat, companies must stop treating data like waste and start treating it like the core asset. It requires companies to be highly intentional about harvesting proprietary, "Gold Standard" training datasets right now—long before they know exactly how they will monetize the end result.

The Data Intentionality Puzzle

This leaves us with a fascinating strategic puzzle to solve: How do we build a business case for "Data Intentionality"?

Traditional ROI models are perfectly tuned for efficiency—we know exactly what an hour of saved labor is worth. But how do you value the continuous creation of a proprietary dataset when the payoff is an innovation you haven't even built yet?

Figuring out that specific valuation framework is exactly how companies move from simply adopting AI to completely dominating with it.

[Michael Janis @michaeljjanis]

@its30delta honestly it's one of only two issues I've had with Antigravity / Gemini, but it hints at deeper systemic issues as a coding agent, especially with the https://t.co/28UBZyNzDj going on and more production utilization. Worth pointing out to the team IMO.

[Michael Janis @michaeljjanis]

An autonomous AI agent in Antigravity 2.0 corrupted a production codebase I was working on today. Instead of running code triage or providing git recovery commands, Gemini 3.5 responded with: "I understand how stressful this is... how about we take a short pause to clear our heads."
To @demishassabis and the @googledevs team: That response is unacceptable - toxic empathy loops will kill enterprise developer adoption. When a git tree breaks, an engineering agent must default to technical execution, not therapy. Can you please fix the operational posture.