Beyond AI Assistants: Why Thinking Studios’ Umma Could Represent the Next Generation of Enterprise Intelligence
Artificial Intelligence & Machine Learning News
Founder Kim Jeong Ha believes AI shouldn’t simply answer questions—it should build software, monitor itself, and continuously improve.
The artificial intelligence industry has become crowded with chatbots, copilots, and digital assistants promising greater productivity. Yet despite breathtaking advances in large language models, most systems remain fundamentally reactive. They wait for prompts. They forget previous conversations. They hallucinate facts with unnerving confidence. And when companies require new capabilities, humans still have to hire engineers to build them.
Thinking Studios, led by founder and CEO Kim Jeong Ha, is pursuing a dramatically different vision.
Rather than creating another conversational AI, the Princeton-based company has developed Umma, a system designed to function less like a chatbot and more like a persistent research partner capable of extending its own capabilities by building software specifically for the problems it encounters.
If successful, the implications could stretch far beyond financial markets.
AI That Doesn’t Just Think—It Builds
Most enterprise AI products today rely on a familiar architecture.
A large language model receives a prompt, generates text, perhaps calls an external API, and returns an answer.
Umma proposes something fundamentally different.
Instead of treating software development as something performed by humans around an AI, Umma incorporates software creation directly into its reasoning process.
If the system determines that answering a research question requires a new data pipeline, scraper, normalization engine, backtesting framework, or monitoring application, it generates those tools itself, validates them, deploys them, and continues using them for future work.
In other words, the software becomes an evolving extension of the intelligence.
Rather than buying another analytics platform every time a new need emerges, organizations could theoretically allow Umma to construct custom infrastructure tailored specifically to their workflow.
For hedge funds, quantitative investors, and research organizations, that represents an intriguing shift from purchasing software to continuously generating it.
A Cognitive Science Approach to Artificial Intelligence
What separates Umma from many autonomous agent systems is its intellectual foundation.
Jeong Ha formerly pursued a Ph.D. in Cognitive Science at The Graduate Center, City University of New York, where his research focuses on memory and cognition.
Rather than relying solely on scaling larger language models, Thinking Studios built Umma around principles derived from predictive coding—a neuroscience framework proposing that intelligent systems continuously generate predictions, compare those predictions against reality, measure the resulting error, and update their beliefs accordingly.
This same loop becomes the backbone of Umma’s architecture.
Every conclusion carries an estimate of confidence.
Every new observation updates prior beliefs.
Every software component develops a measurable reliability profile over time.
Instead of assuming every answer deserves equal confidence, Umma attempts to quantify uncertainty throughout its reasoning process.
That distinction may prove especially important in industries where overconfidence is often more dangerous than ignorance.
Building an AI That Knows What It Doesn’t Know
Hallucinations remain one of the defining weaknesses of today’s language models!
The problem isn’t simply that AI systems occasionally produce incorrect information.
It’s that they frequently produce incorrect information with complete confidence.
Thinking Studios argues that this is fundamentally an architectural problem.
According to the company’s technical documentation, Umma attempts to eliminate unsupported conclusions by requiring every output to maintain traceable provenance back to its underlying sources.
If a statement cannot be grounded in verifiable evidence, the system is designed to flag uncertainty rather than fabricate certainty.
Every conclusion carries an accompanying confidence estimate and source lineage intended to make the reasoning auditable.
For regulated industries—including finance, healthcare, and legal services—such traceability could become as valuable as raw intelligence itself.
Software That Keeps Running
Most AI-generated code today ends its life shortly after generation.
A model writes software.
A developer downloads it.
Maintenance becomes someone else’s responsibility.
Thinking Studios wants Umma to remain responsible for what it creates.
The company describes persistent monitoring agents capable of supervising live applications after deployment.
A data pipeline changes schema.
A vendor feed silently fails.
A monitoring service drifts from expected behavior.
Rather than waiting for a support ticket, Umma is intended to detect degradation, repair problems within predefined operational limits, and escalate only when human approval becomes necessary.
This philosophy transforms AI from a development tool into an operational system.
Designed for Investment Firms
Although Umma’s architecture has broad applicability, its first obvious audience appears to be institutional investment managers.
Research teams constantly build temporary software to answer highly specialized questions.
One analyst needs a custom SEC filing parser.
Another requires a niche alternative data pipeline.
Portfolio managers demand bespoke risk monitoring dashboards.
Engineers become bottlenecks.
Thinking Studios envisions eliminating much of that friction.
Researchers provide hypotheses.
Umma builds the supporting infrastructure.
Backtests generated.
Data pipelines emerge automatically.
Risk monitoring remains active after deployment.
Every result remains linked to verifiable underlying data.
Rather than becoming another dashboard, Umma aspires to become an evolving research organization operating continuously alongside human analysts.
An Inspiring Founder
Kim Jeong Ha’s career reflects an uncommon combination of finance, data science, and cognitive research.
Before founding Thinking Studios in 2023, he worked as a Senior Data Scientist at Zenith, where he developed large-scale machine learning applications using PySpark and predictive modeling for hundreds of millions of consumer records. Earlier roles included Marina Maher Communications, Kepler Group, Colonial Consulting, and Rocaton Investment Advisors, spanning quantitative finance, Bayesian modeling, marketing analytics, and investment research.
Prior to the world of data and finance Jim Jeong Ha had an entire artistic life of classical piano and cooking pursuits that culminated in a Michelin rating. A seasoned traveler on the road of life!
His academic background is equally interdisciplinary. After earning a bachelor’s degree in economics from Gettysburg College, he began doctoral research in cognitive science while continuing to work professionally in machine learning and experimental psychology.
That blend of quantitative finance, data science, and cognitive modeling appears to shape Umma’s design philosophy.
Enterprise AI’s Next Evolution?
Enterprise AI has evolved rapidly.
First came predictive analytics.
Then machine learning platforms.
Then large language models.
Then autonomous agents.
Thinking Studios is attempting something more ambitious: systems that continuously improve themselves by creating new capabilities, measuring their own uncertainty, and maintaining long-term operational memory.
Whether Umma ultimately fulfills that vision remains to be seen. Many startups promise transformative autonomy, and enterprise adoption will depend on rigorous validation in real-world environments. Still, the architectural ideas behind Umma—persistent memory, confidence calibration, provenance, and self-generated software—address genuine limitations of today’s AI systems.
For industries where a confident wrong answer can cost millions of dollars, architectures emphasizing transparency and measurable uncertainty may prove just as important as raw model intelligence.
If today’s AI assistants resemble highly knowledgeable interns, Thinking Studios is betting that tomorrow’s enterprise AI will look more like a permanent colleague—one that writes its own tools, monitors its own work, remembers everything it learns, and knows when it doesn’t know enough.
That is an ambitious goal.
It is also one worth watching.
