January 2026

The Simulation
Worldview

Every Silicon Valley startup pitches a simulated, modeled future. The map has replaced the territory.

The Pitch

Every pitch deck tells the same story. We will model your business. Predict your customers. Optimize your operations. Simulate every outcome before it happens.

94%of enterprise software promises "predictive insights"

78%of AI startups claim to "model reality"

The assumption runs deep. If we can measure it, we can model it. If we can model it, we can predict it. If we can predict it, we can control it.

The map is not the territory. The model is not the system.

The Failures

0major recessions predicted by economic models

7 daysmaximum reliable weather prediction window

0cultural shifts predicted by social models

$2.3Tlost to model-based financial strategies in 2008

The world resists modeling. It resists prediction. It resists the clean lines of a simulation. The best weather models fail a week out. Economic models miss every recession. Social models fail to anticipate any cultural shift that matters.

And yet we keep building products that assume the model IS the reality. That treat the dashboard as truth. That mistake correlation for causation. That confuse precision with accuracy.

The Problem

The simulation worldview creates products that work brilliantly in demos and fail catastrophically in reality. They optimize for the modeled world, not the actual one.

Every model is a simplification. Every simplification is a loss. Every loss matters in ways we cannot predict.

The hubris is not in building models. Models are useful. The hubris is in forgetting they are incomplete. In treating them as oracles rather than approximations.

The Pattern

The same story plays out across industries. A model works in testing. It gets deployed. Reality diverges from the model. The system fails. Everyone is surprised.

Financial Models

Black-Scholes assumed markets were rational. They built trillion-dollar positions on this assumption. When markets stopped being rational, they collapsed.

Recommendation Engines

Optimized for engagement metrics. Created filter bubbles, radicalization pipelines, and mental health crises. The model worked. The outcome was disaster.

Self-Driving Cars

Trained on millions of miles. Still fail at scenarios no simulation anticipated. An edge case to a model is just Tuesday to reality.

Pandemic Models

Predicted curves and peaks with precision. Missed human behavior, political will, supply chains, and everything else that mattered.

The simulation worldview isn't wrong because simulations are useless. It's wrong because it forgets they're incomplete.

Build for variance.
Not just means.

What Good Looks Like

The best products know what they do not know. They build for the unexpected. They design for failure modes, not just success paths. They respect the complexity they cannot capture.

They treat models as tools, not truths. As starting points, not destinations. As maps that are always, necessarily, incomplete.

We believe:

1Models are hypotheses. They should be tested, not trusted. Validated, not venerated.
2Edge cases are not edge cases. They are the moments when your product meets reality.
3Graceful degradation is required. When the model fails, the product should not.
4Human override is non-negotiable. No system should be trusted more than the people using it.

What We Build

Robust Systems

Products that work when assumptions fail. Designed for the unexpected, not just the modeled.

Transparent Predictions

AI that shows its uncertainty. Models that communicate their limitations.

Human-Centered Automation

Systems that augment human judgment, not replace it. Tools, not oracles.

Reality-Tested Products

Built for actual conditions, not simulated ones. Validated in the wild.

Is your product designed for the model?

Or for reality?

Written by · January 6, 2026

Donny Smith · ECD, Founder, Bttr.

Over the past 15 years, he has led creative teams and contributed to products used by millions of people worldwide, working with companies including Apple, Opendoor, JP Morgan, GE Aerospace, Pepsi, and Alterra Mountain Company.

Share this perspective

More Insights

Perspectives on product design, technology, and the forces shaping how people interact with the things we build.

8 min read

How to Evaluate a RAG System

Evaluating a RAG system means scoring its two stages on their own, because a retrieval augmented generation answer is only as good as the passages behind it, and a wrong answer has two possible causes a single number cannot tell apart. Retrieval is measured before the model writes anything. The LlamaIndex evaluation guide scores a retriever with hit rate, the share of questions where the right passage lands in the top results, and mean reciprocal rank, which rewards ranking it high. Generation is measured given what was retrieved. Ragas, the open evaluation toolkit now maintained by VibrantLabs, splits the answer into claims and scores faithfulness as the share the retrieved context supports, then scores answer relevance by generating questions from the answer and comparing them to the real one. Read as two pairs, the four metrics locate the fault: high recall with low faithfulness is a generation problem, low recall with high faithfulness is a retrieval one. Anthropic showed the payoff in September 2024, cutting a retrieval failure rate measured by recall at 20 from 5.7 percent to 1.9 percent by adding context, keyword search, and reranking. And it is not a one time check. Models drift, documents go stale, so the teams that trust their systems run these scores on live traffic the way they run unit tests.

8 min read

What Is a Reasoning Model

A reasoning model is a large language model trained to work through a problem in steps before it answers, rather than reply in one pass. It spends extra computation at the moment you ask, called test-time compute, to write out a chain of thought and check its own work. Anthropic describes its extended thinking as letting the very same model give itself more time and effort rather than switching to a separate model, and reported that Claude 3.7 Sonnet reached 84.8% on the hard GPQA science set, with a 96.5% physics subscore, using parallel sampling at a 64K token thinking budget. The habit is trained, mostly by reward. DeepSeek showed in January 2025 that its R1 Zero model learned to reason purely through reinforcement learning with no supervised fine-tuning first, developing self checking and long chains of thought on its own, and shipped DeepSeek R1 under the MIT license at performance comparable to OpenAI o1. The cost is real, so the newest models add a dial. Claude Opus 4.5 hit its best score on SWE-bench Verified using 76% fewer output tokens at a medium effort. The skill is matching how hard the model thinks to how hard the task is.

8 min read

What Is Fine-Tuning

Fine-tuning is training a pre-trained model further on your own labeled examples so a behavior is set into its weights. It changes how a model acts, its tone, its format, its grip on a narrow task, rather than what facts it knows. Google frames the move as taking a general-purpose model and adapting it to your domain on a curated dataset of your own examples. The gains are measurable. When Anthropic opened fine-tuning for Claude 3 Haiku in Amazon Bedrock, a team fine-tuned it to moderate comments and lifted classification accuracy from 81.5% to 99.6% while cutting tokens per query by 85%. Lighter methods such as LoRA keep the base model frozen and train a small set of new weights beside it. The order that saves money is a plain prompt first, retrieval when the model lacks facts, fine-tuning only when it lacks the right behavior. A fact that changes is a retrieval problem. A behavior that drifts is a training problem.

View all insights

The companies that build for reality instead of simulation will own the markets that matter. The ones where failure has consequences.

The Simulation
Worldview

The Failures

The Problem

The Pattern

Build for variance.
Not just means.