We are thrilled to announce Radical Ventures’ lead investment in Mosaic‘s $18M Series A round as the company builds the AI-native operating system for private markets investing including private equity, credit and investment banking.
GenAI has unlocked an endless number of use cases that are providing real value in enterprises today. But LLMs are probabilistic, and still struggle with structured, tabular data. They are sometimes correct, sometimes wrong, and often nearly impossible to audit.
For the highest stakes use cases, “sometimes correct” isn’t good enough. The Mosaic team has built what we think of as the “trust layer” for AI in the private markets. With bulletproof and auditable math, Mosaic has become the go-to financial modelling solution for five of the top ten (and counting) global private equity funds and bulge bracket banks.
AI will not replace human judgment, but it is quickly eliminating wasted human labour. Mosaic enables deal teams to build IC-ready models in minutes rather than days and guarantees defect-free models (no linking errors, busted formulas, or rogue hard-codes).
Mosaic’s founder, Ian, spent a decade in private equity before founding the company, and is a self-taught coder who wrote the first version of the platform himself. That rare combination of deep domain expertise and technical fluency is what has allowed Mosaic to create and hold the category of AI-native deal modelling, when most in the industry thought Excel was unassailable.
Gone are the days of checking models cell by cell or waiting 72 hours to see a revised operating case. In a world where speed and certainty win deals, this allows investors to spend more time on the pieces of diligence that matter and move faster than their competition.
The best investors and bankers are differentiated by their judgment, networks, and strategies, not the number of keystrokes in Excel. We’re excited to be a part of that transformation.
Read more in Mosaic’s announcement of the round.
AI News This Week
-
Cohere and Aleph Alpha Join Forces to Form Global AI Powerhouse (Press Release)
Radical Ventures portfolio company Cohere has agreed to acquire German AI lab Aleph Alpha in a deal supported by the German and Canadian governments, who are expected to be anchor customers. The combined entity will have its global headquarters in Toronto and its European headquarters in Berlin. The deal will give Cohere greater access to the European market, where companies, governments and public institutions are seeking alternatives to American tech as part of a broader digital sovereignty push. As Cohere’s CEO and co-founder, Aidan Gomez put it: “We are uniting under the Cohere brand to create a global and independent AI powerhouse.” Schwarz, the €175 Billion/year German industrial colossus, will invest $600 Million as part of a new funding round in Cohere.
-
Yutori Launches Delegate to Turn AI Agents into Proactive Web Workers (Silicon Angle)
Radical Ventures portfolio company Yutori has launched Delegate, a product that lets users hand off everyday digital tasks to an AI agent that completes them autonomously. Delegate marks the progression from Yutori’s earlier releases, Scouts for web monitoring and Navigator for vision-based web page interaction, into a full end-to-end task execution platform. Rather than simply responding to prompts, Delegate researches the problem, builds a structured plan, and connects to external tools to carry out the work without further user involvement.
-
10 Things That Matter in AI (MIT Technology Review)
Several Radical portfolio companies are at the forefront of the innovations shaping the AI market right now. Cohere and Reka are advancing the push toward cheaper, more efficient LLMs through enterprise-grade models, while Yutori‘s Delegate takes an agent-first approach to autonomous task execution as multi-agent orchestration moves beyond coding into general workflows. Periodic Labs is building a closed-loop AI co-scientist that pairs specialized agents with robotic laboratories for materials discovery, and World Labs is constructing 3D world models to give AI systems a grasp of the physical environment.
-
Why Your AI assistant is Suddenly Selling to You (Economist)
Advertising is migrating into AI chatbots as AI labs seek to monetize free-tier LLM inference. Some platforms front-load ads like a conventional search engine, while others wait up to ten conversational turns before serving one, betting that intent becomes clearer over time. Radical portfolio company Firsthand is building an alternative model that enables brands and publishers to deploy their own AI agents to engage consumers directly with proprietary knowledge, giving companies control over how their expertise reaches audiences.
-
Research: Memory Transfer Learning in Coding Agents (KAIST/NYU)
Coding agents today build memory from past tasks but typically only reuse it within the same domain. This paper tests whether memories from unrelated coding tasks can transfer across domains, and finds that a shared memory pool drawn from six different benchmarks improves average agent performance by 3.7%. The gains come almost entirely from meta-knowledge, things like validation routines and structured edit-test workflows, rather than task-specific code. High-level insights generalize well across domains, while raw execution traces often hurt performance by anchoring agents to irrelevant implementation details.
Radical Reads is edited by Ebin Tomy (Analyst, Radical Ventures)
