Anthropic and Google both had massive dev days recently. And they couldn't be more different. In this episode, Jhanvi and Michael break down what each announcement signals about the frontier model race and where it's headed. Anthropic is doubling down on enterprise agents, memory stores, and "dreaming," while Google is going wide with consumer AI, a multimodal Omni model, and Spark embedded across its entire product suite.

They also get into a question that comes up with clients and candidates alike: how worried should companies actually be about vendor lock-in? Plus: what happens when you run the same agentic harness with different frontier models, why tokens per second is becoming a more important metric, and why you shouldn't switch back and forth between Cowork and ChatGPT.

Key Takeaways

• Decouple Architecture via Open Standards: To prevent long-term vendor lock-in, firms should deploy custom skill libraries and organizational knowledge layers as open, text-based formats stored in client-owned GitHub repositories rather than within proprietary model environments.

• Implement OpenTelemetry Early: The highest hurdle to switching model providers is the loss of historical session data; setting up an independent OpenTelemetry system up front ensures your firm owns its telemetry and interaction data, permitting smooth cross-provider migration.

• Isolate Compute with Managed Sandboxes: Utilizing self-hosted agent tool containers allows institutional firms to keep localized data execution and tools within their secure cloud environments while securely executing the core inference loop via external APIs.

• Focus on Immediate ROI Over Early Optimization: Many firms stall their AI adoption by over-engineering cross-cloud or cross-vendor compatibility too early. Successful deployment requires mastering one ecosystem to capture immediate time-to-value before optimizing for compute spend arbitrage.

About Hedgineer

Hedgineer is building the AI platform for institutional investing — deploying agents, skills, and data connectors directly inside hedge funds and asset managers to transform investment and operational workflows.

The Hedgineer Podcast follows CEO Michael Watson and COO Jhanvi Virani as they navigate the frontier of AI adoption in finance, sharing unfiltered perspectives from the teams, guests, and problems they work with every day.

Subscribe for weekly analysis on AI infrastructure and institutional finance.

Watch the full episode on Spotify or YouTube at youtube.com/@hedgineer.

Connect with us on LinkedIn at linkedin.com/company/hedgineer-io or reach out at podcast@hedgineer.io.

Hedgineer.io

The transition from AI as a chatbot to AI as an autonomous agent requires more than just better models; it requires agents capable of regulating their own state and context at scale.

In this episode of The Hedgineer Podcast, co-hosts Michael Watson and Jhanvi Virani sit down with Mitch Troyanovsky, co-founder of Basis, an agent platform specifically designed for the accounting industry. The conversation moves beyond the hype of generative AI to address the engineering realities of building "agent-native" enterprises. Mitch explains why the next frontier of applied machine learning involves closing the loop on self-improving agents—systems that can optimize their own trajectories, contexts, and tools without constant human intervention.

We explore the "single pane of glass" debate: whether specialized platforms like Basis will remain the system of record or if frontier model interfaces will eventually consolidate all enterprise workflows. The discussion delves into the technical nuances of Recursive Language Models (RLMs) and the "Better Intelligence" approach, where models are leveraged to programmatically curate their own context windows to maintain performance over long-duration tasks.

The episode also tackles the cultural shift required for AI adoption. From implementing "Do You Stand By This" (DYSB) protocols to ensure accountability, to the "lexical taxonomy" required to write documentation specifically for LLM consumption rather than human readers, we provide a blueprint for firms looking to move from experimental AI to production-grade agentic systems.

Key Takeaways:

• Closing the Applied ML Loop: Why the next generation of agents will focus on self-regulation and autonomous state management to handle production workloads.

• The "Database-ification" of SaaS: How AI agents interacting via API threaten the value proposition of traditional software UIs, potentially reducing many SaaS tools to mere structured data stores.

• Recursive Language Models (RLMs): A technical look at using model intelligence to dynamically curate context at every forward pass, moving beyond simple "append-only" context windows.

• Writing for Machines: Why traditional human writing styles are inefficient for LLMs and how "information density" is becoming a critical engineering discipline.

About the Guest:

Mitchell Troyanovsky is the co-founder of Basis, a New York-based platform building AI agents for the accounting industry. He is a leading voice on the future of agentic systems at scale and the implementation of Recursive Language Models in production.

About Hedgineer

Hedgineer is building the AI platform for institutional investing — deploying agents, skills, and data connectors directly inside hedge funds and asset managers to transform investment and operational workflows.

The Hedgineer Podcast follows CEO Michael Watson and COO Jhanvi Virani as they navigate the frontier of AI adoption in finance, sharing unfiltered perspectives from the teams, guests, and problems they work with every day.

Subscribe for weekly analysis on AI infrastructure and institutional finance.

Watch the full episode on Spotify or YouTube at youtube.com/@hedgineer.

Connect with us on LinkedIn at linkedin.com/company/hedgineer-io or reach out at podcast@hedgineer.io.

Hedgineer.io

Anthropic just announced an enterprise services venture backed by Goldman, Hellman & Friedman, and Blackstone. OpenAI is raising $4B for something similar. So why are frontier model providers suddenly trying to become consultants?

In Season 3 Episode 3 of The Hedgineer Podcast, Michael and Jhanvi break down what's driving the move: why handing a company a Claude license rarely translates into real automation, and why building domain-specific is critical to successful deployments. The real unlock behind all of this is agent harnesses, which have expanded what AI can do far beyond a chat interface. They dig into how providers are approaching harnesses differently and why state management and organizational memory are the differentiators that not enough people are talking about.

Plus: GPT 4.5 vs. 5.5 cost dynamics, why understanding model caching could save your company thousands of dollars, and whether Apple is sitting on the consumer unlock that could shift public skepticism on AI.

About Hedgineer

Hedgineer is building the AI platform for institutional investing — deploying agents, skills, and data connectors directly inside hedge funds and asset managers to transform investment and operational workflows.

The Hedgineer Podcast follows CEO Michael Watson and COO Jhanvi Virani as they navigate the frontier of AI adoption in finance, sharing unfiltered perspectives from the teams, guests, and problems they work with every day.

Subscribe for weekly analysis on AI infrastructure and institutional finance.

Watch the full episode on Spotify or YouTube at youtube.com/@hedgineer.

Connect with us on LinkedIn at linkedin.com/company/hedgineer-io or reach out at podcast@hedgineer.io.

Hedgineer.io