Three years ago, the Valley was obsessed with "Scaling Laws." The belief was simple: feed the beast more tokens, and AGI emerges.

That thesis has hit a wall.

Look at the evidence. We have LLMs that pass the Bar Exam but fail the "Reversal Curse" (knowing A=B, but failing to deduce B=A). We see "Hallucination Loops" in RAG systems that confidently lie about financial data because they are probabilistically associated, not causally orchestrated. The industry is stuck on Judea Pearl’s Ladder 1: Association (Seeing). They are building faster horses. We are building the Intelligent Causal Machine.

As we enter 2026, I want to anchor us on a single question. Not what we ship this quarter, but who we are at our core.

Who are we?

We are an AI company headquartered in Silicon Valley.

We build the most powerful Causal AI for all.

Causality as "System 2" (Product)

The frontier is no longer "Chat"; it is "Reasoning."

Everyone from OpenAI to Google is trying to solve "Reasoning" (System 2). But they are doing it by chain-of-thought prompting on top of statistical models. It’s inefficient and opaque. Yann LeCun was right: “Autoregressive LLMs cannot plan and reason about the world.”

We don’t just predict text. We model Causality to interconnect the world’s states, facts, and data.

The Difference:

• The Collector (Ladder 1): A user asks, "Where can I find the Labubu 'Zimomo' limited edition?"
  
  

  • Competitor: Lists 10 eBay links. "Prices are trending up."

  • Abel: We analyze the Scarcity Flow. "Supply is artificially high because the Tokyo Pop Mart Expo is this week. If you wait 14 days, the post-expo dump will cause prices to drop by 20%. However, if you intervene and buy the only two 'Secret' editions on Xianyu right now, you effectively corner the regional supply, forcing the local price up by 40%."
    
    

• The Prediction (Ladder 1 vs 3): A user asks, "Will the Fed cut rates in Feb?"
  
  

  • Competitor: "75% chance based on consensus and Twitter sentiment."

  • Abel: We model the Legislative Influence Graph. We highlight a Critical Divergence: The Committee's decision function is causally locked to Supercore Services Inflation, not headline CPI. That specific node hasn't moved. "Abel calculates the true probability is 12%. The market is hallucinating a pivot because it looks at the signal (prices); we see the blocker (mandate). If you intervene and assume inflation drops 30%, the probability flips to 70%."

Knowing why—and being able to intervene—is always more powerful.

AI + Data, but data always comes first

Data is not just tokens; it is fact, number, graph and actionable insights.

In 2024-2025, companies like Reddit and Stack Overflow sold their data for millions just to train LLMs to talk better. That gold rush is over. The new oil is Causal Graphs and Causal Models.

The winners in Biotech aren't just reading papers; they are running wet-lab experiments to find causal links between genes and diseases. The winners in finance aren't just analyzing historical charts; they are simulating Liquidity Cascades to predict how a single margin call in Tokyo causes a butterfly-crash in New York, long before the ticker moves. The winners in E-commerce aren't just recommending "similar items"; they are running Counterfactual Pricing models to prove that a specific $5 coupon didn't just "correlate" with a purchase, but was the sole causal mechanism that prevented a high-value user from churning to a competitor. The winners in Information aren't just indexing keywords; they are mapping hidden facts to show you not just what the news is, but why the narrative shifted—tracing the causal root of a trend back to a single lobbyist’s whitepaper rather than just summarizing the noise on Twitter.

Then, what makes it the Abel Way? Data. We don't just ingest a company's PDFs or some random text to feed into LLM to ramble. We ingest their logs, their decisions, their outcomes and numbers to build a Causal Graph of the hidden mechanism flow. We own the "Causal Insight Layer," not just the "Raw Data Layer".

If we have continuous variables—millions of rows of server logs, tick prices, or sensor telemetry—we unleash Large-scale Causal Discovery (CD) algorithms to mathematically prove the hidden structure.

If we have discontinuous variables or messy natural language, we don't just chat with it. We weaponize LLMs as the general Causal Discovery Compute Engine to extract and verify the causal edges that math alone might miss, then collect more data to further verify these edges and nodes by large scale CD algorithms.

Competitors see a data lake; we see a refinery. We take the sand of raw data and turn it into the oil of causality driven strategy. We graze on the grass of unstructured text and process it into the milk of reasoning logic behind the scene. While they hoard bytes, we refine wisdom.

CGI infrastructure + Ecosystem

Trust and authenticity is the new speed.

Why is the world still hesitant to let AI take the wheel? Because current AI is a black box without authenticity. If an LLM suggests a geopolitical strategy or a climate policy, it says, "Trust me, the probability distribution aligns." That is not enough for the real world. That is liability. That is guessing.

Abel provides the "Why Trace." We are the World Simulator. Because our core is the World Causal Model on the World Causal Graph, every output comes with a visible, auditable causal path. We don't just give the answer; we give the mathematical proof of the consequence. This rigor allows us to scale from analyzing data to simulating reality.

From "Indexing" to "Simulating"

We are rewriting the internet's fundamental behavior verb. The history of the digital age is defined by how we interact with information - 2010: "Google it" (Search) You needed to find the information. Google indexed the links. You did the reading; 2023: "GPT it" (Generate and Synthesize) You needed to summarize the information. LLMs read the internet and give you a blurb. But they couldn't tell you if it was true, or what would happen next; 2026+: "Abel it" (Simulate) You need to solve the reality. You don't want a summary of the past; you want a simulation of the future.

We want the world’s smartest developers to stop building flimsy chatbots that hallucinate. We want them to build robust Intelligent Agents on top of our CGI (Causal Graph Intelligence) protocol. When a developer builds on OpenAI, they get a parrot. When they build on Abel**, they inherit our causal logic, our authenticity, and our foresight.** We enable builders to reason causally, products to adapt via intervention, organizations to simulate futures before they commit to them.

This is our ecosystem play. We are building the World Causal Engine for Reality.

Our Causal Taste

This Is Non-Negotiable.

We need to stop looking at OpenAI or Anthropic for design cues. Stop copying the chat bubbles. Stop copying the bland "Corporate Memphis" art style. Stop trying to look "friendly" but similar to 100 other apps.

Everything we do - Code, Design, Tech, Marketing, Sales—must drip with our distinct Causal Taste and Abel Personality.

Here is what that looks like across the company:

Experience-Wide Fun

• Kill the Chat Bubble: Chat is for texting your friends. It is a passive interface. Abel is an active engine (we push first rather than pull).

• The Aesthetic: We are "God-Mode" software. Think Minecraft meets Bloomberg Terminal, but designed by Apple.

• Tactile Causality: When a user intervenes in the system, it shouldn't just print text.

  • Give me sliders, bars and numbers. Give me nodes that pulse when connected.

  • When I change a variable, I want to see the ripple effect tear through the graph in real-time.

  • It should feel like playing SimCity or Civilization. It should be addictive and sharp. Making a decision should feel like launching a rocket—precise, mechanical, and satisfying.
    
    

Technology: Technically Sharp, AI Deep

• Always show the magic, causality is a modality: Competitors hide the complexity. We visualize the complexity.

• Our UI must reflect our code. If the backend calculates a confidence interval, the frontend must show the error bars or the big number tractions. If the backend has the weights and bias, the frontend should have playable buttons for call-to-action (CTA).

• Respect the Math: Treat us mathematically educated, don't dumb it down. If the Causal Discovery algorithm finds a Collider Bias, use it to calculate. If we have priors or assumptions produced by causal graphs or even LLM prompts, adopt it in belief propagation. LLM \+ Mathematical Causal Engine \+ LLM as a sandwich is great, but only LLM can bring us to nowhere.

• Causal AI Deep: We use LLMs, but we keep them on a leash. The LLM is the translator; the Causal Engine is the truth. Never let the translator override the truth.
  
  

Brand, Design & GTM: Never Follow

• The Voice: We are not "Helpful & Cute." We are "Certain & Sharp."

• The Narrative:

  • They sell Magic. (Magic is fake).

  • We sell Math and Physics. (Math is real!).

• Visuals: Be Unique, opinionated but also user-adoptable. Use Diagrams. Use Schematics. Use Numbers. Use Highlights.

• The Vibe: We are the engineers who walked into the room to fix the mess by making mathematical poets. We offer clarity, structure and fun in a world of fake hallucination and inauthenticity.
  
  

One more thing - Causality First

We will not let LLMs, prompts, context engineering, or those superficial trends define us.

At Abel, LLMs are components, Math and causal AI are the frame. Correlation is a baseline, causal drivers and actionable outcome is the goal.

We design everything using Causal AI methodology. We have Causal Graphs, exploit it at every corner in our PRD, Design, UI/UX and marketing. Why? OpenAI doesn’t have it, Athtropic doesn’t have it, Google and Facebook doesn’t have it, Deepseek and Bytedance doesn’t have it, then what are we waiting for?

If we don’t have the calculated Causal Graph and Causal Model, is it the end of the game? No, we can deploy knowledge and LLM to put reasonable prior, calibrate the number, verify the node and edges across corpus, context and modality.

Then we do the math\! Deploy belief propagation at scale, use Linear Solver or residual algebra, deploy large scale Double Machine Learning by matrix calculation. Others optimize correlation at Ladder 1\. We climb Ladders 2 and 3—and we build products, OS and infra so others can climb with us.

Abel is not here to follow the GenAI or LLM wave. We are here to change how intelligence is built.

If this excites you, you’re in the right place. Let’s make causality unavoidable.

Onwards and upwards,

~Stephen & Biwei