GOVERNANCE

 AI firms court finance professionals

AI firms have been fighting for enterprise attention and dollars. Now, they’ve got a new target: Finance. 

Several major AI firms have launched new tool initiatives aimed at finance professionals. Seeking to embed themselves and scale within a particularly lucrative market, these tools come as both OpenAI and Anthropic aim to boost revenue and race towards profitability ahead of their public offerings. 

Here’s what was announced:

• Claude financial service agents: Anthropic released ready-to-run agent templates for financial services tasks, including building pitch books, screening KYC files, reviewing valuations and closing books. Claude also now works across Microsoft Excel, PowerPoint, Word and soon Outlook through the Claude add-ins for Microsoft 365. 

• OpenAI and PwC’s Native Finance Function: The ChatGPT maker and professional services firm announced a collaboration to build agents around the “core operating rhythms” of finance, including planning, forecasting, payments, treasury, taxes and accounting. The organizations noted that the partnership is already in progress, building a procurement agent inside OpenAI’s finance organization. 

• Perplexity Computer for Financial Services: The self-described AI answer engine has extended the reach of its “general-purpose digital worker” to finance. Now, finance teams can bring licensed data from providers like Morningstar and Pitchbook into the agent to work with 35 dedicated workflows for tedious analyst work.

While the upside is obvious for AI firms, these tools also present real potential to democratize financial analysis, Terra Higginson, a principal research director at Info-Tech Research Group, told The Deep View. Until recently, AI has been “notoriously bad at financial analysis.” But with access to proper data, these tools have improved vastly, she said. Now, private investors could have access to tools that would have previously only been available to the top tier of investors and institutions. 

“I still would not blindly use any model for financial analysis … but this is the first time the direction feels usable as a tool and not just a huge risk,” Higginson said. 

However, risks still exist. Along with the obvious data security and regulatory risks, if everyone is using the same models, leveraging the same data, and coming to the same conclusions, “markets could become more crowded, more reactive, and potentially more volatile,” Higginson said. The real edge isn’t in trusting outputs outright, she said, but rather sharpening human judgment and knowing when not to trust the model. 

“We always need to keep a human in the loop. I cannot emphasize this enough,” Higginson said. “Finance is highly regulated, high-stakes, and way too risky to let models operate alone.”

The problem with implementing AI into any risky industry is always speed. Now more than ever, executives are feeling the pressure to get returns from their models, with many CEOs feeling their job is at stake if they don’t get it right as soon as possible. That pressure could easily lend itself faster, and potentially sloppier, deployments. In financial contexts, that sloppiness could have dire domino effects. So while these tools could completely transform the world of finance, enterprises, investors and professionals may want to tread carefully. 

TOGETHER WITH GRANOLA

The Deep View team is obsessed with Granola.

We're not talking about the food (although a few of us have an unhealthy interest in that too).

We're talking about the AI notepad we've been using in 2026. It works across our teams, summarizes every meeting, and saves us around 10 hours a week per person.

Try it free for a month for your team with code THEDEEPVIEW.

PRODUCTS

Teradata wants to tidy up enterprise AI chaos

AI and the data to leverage it can often feel like a scattered mess for enterprises. Teradata wants to help them clean house. 

On Wednesday, Teradata unveiled the Teradata Autonomous Knowledge Platform, its new flagship product meant to unify different aspects of AI systems, including production-ready AI models, structured and unstructured data, and analytics, into a single system for “trusted, governed understanding.” 

In practice, this platform is meant to give customers more insight into the cost and performance of AI systems with appropriate governance in place, giving organizations better control over their models and data — fulfilling a need that has only grown with the advent of AI agents. The platform spans across cloud, on-premise, and hybrid environments.

“Grounded in industry-specific data, semantics, and lineage, it provides the business context for agentic AI to sense, decide, and act reliably and repeatedly across systems and tools — with minimal human intervention — while learning and improving over time,” the company explains in the release. 

The platform is also meant to help enterprises move from the pilot to production phase by giving companies clear insights into their AI operations to better inform decisions around deployment and improve outcomes.

“Enterprises moving fastest are already driving 10× gains in speed, cost, and productivity," said Ray Wang, analyst at Constellation Research, in the release. “The ones falling behind are still running pilots. Breaking that cycle means a strong data foundation, outcomes-based AI, and real governance as a single system, not assembled from parts.” 

The platform’s other new capabilities include:

• Teradata AI studio: An environment where all AI capabilities can be found and can be used for agentic analytics, model lifecycle management, etc. It is also available separately for organizations that want to build it into existing infrastructure.

• Tera Agents: Pre-built platform agents that can help with tasks such as cost optimization, including a sizing agent, telemetry agent, FinOps agent, tuning agent, and compute agent.

• Tera: Teradata's autonomous AI workspace that transforms natural language into enterprise-scale data intelligence.

The Teradata Autonomous Knowledge Platform is expected to be available in the third quarter of this year on Teradata Cloud, but Teradata AI Services and AI Studio are available now.

AI’s rapid ascent among enterprises has greatly stressed the significance of data, as it ultimately fuels and shapes how these models behave. But the importance of data goes beyond simply what goes into the models, but also their outputs and how they perform. At the enterprise level, it is extremely easy to burn through a lot of tokens (and therefore money) without explicitly seeing results. Ideally, better visibility into the data could help companies better understand what's working and what isn’t. Broadly, understanding this can help with cost and resource efficiency, something that’s vital as demand for AI outgrows the supply of compute. 

TOGETHER WITH VIKTOR

The Agent for Everyone Else

Most AI agents are built for engineers. They live in your IDE, write code, and assume you know how to chain tools.

Viktor is the agent for everyone else. Operations, finance, marketing, sales, support. The people who run the business in Slack.

One message: "draft the Q2 sponsor proposal." "Audit our Klaviyo flows." "Reconcile yesterday's transactions." "Research this lead before my 3pm."

Viktor pulls from your tools, does the work, ships the output. 3,000+ integrations. SOC 2 compliant. No prompt engineering.

Teams on Viktor build what used to take a developer, an analyst, and a week.

Get Started for Free

WORKFORCE

Enterprises are underutilizing AI, says IBM CEO

While it may seem like AI is being adopted by every company, most aren’t  using it to its full potential. 

That was the core message IBM CEO Arvind Krishna expressed in the opening keynote for the company's annual Think conference: "Most enterprises run AI at the margin." Rather than reimagining their core processes, he argued, businesses are settling for incremental improvements at the edges of their workflows.

“But the core, end-to-end processes – and a process is, in the end, how an enterprise makes money, makes revenue, … those are largely untouched in the vast majority,” added Krishna. 

That’s because we are currently in “day zero” of the AI revolution, Krishna said, which  means that the revolution is here and companies have to take advantage of it now. 

Yet, in a separate Q&A with analysts and select members of pressed, he attributed people’s hesitation of adopting it now to a fear of the unknown, such as risk, employee’s reactions, and ROI. But Krishna contests this fear with a simple question: If so many people are sharing success stories, why is your company so special that those use cases wouldn’t work for you? 

Rather, he advises companies to start slow. 

“The world has shown that there is a better way to do something, and if you don't embrace it, it will be a slow oblivion,” said Krishna. “So my advice to them is always pick two or three – I'm not saying pick 10 – areas where it could scale massively, and then, as opposed to doing 100 experiments, pour your energy into making sure that those work.”

He added that those two projects should be chosen only if the  leadership of those operations are interested, not just because of what AI can do, as they will be motivated to work around the roadblocks. Once those projects are deployed, Krishna advised to use the confidence as a jumping off point to take it a step further and accelerate other AI projects.

In my years covering the AI beat, I've spoken with leaders across industries and roles, and nearly every one of them has echoed the same advice as Krishna, to  start small with a few purposeful projects, then scale accordingly. There's a reason this guidance keeps surfacing: it holds up. Among the biggest challenges plaguing the industry today are rising compute costs, elusive ROI on AI initiatives, and the all-too-common trap of getting stuck in the pilot phase, unable to bring projects fully to production. Taking a bite-sized approach is one of the most effective ways to minimize those risks.

Disclosure: Sabrina Ortiz's travel to IBM Think was paid by IBM. The Deep View's coverage is editorially independent from the companies we cover.

LINKS

• Anthropic signed a deal with SpaceX for compute

• Apple to pay iPhone 16, 15 Pro owners $250 million for not delivering Siri 2.0

• Microsoft considers backing away from clean energy targets 

• OpenAI introduces ChatGPT Futures, giving grants for AI-driven research

• DeepSeek in talks to raise up to $4 billion at $50 billion valuation

• Google updates AI overviews to include social media, Reddit posts

• Ads in ChatGPT: OpenAI’s flagship chatbot now has a self-serve ad platform

• Inworld Realtime TTS-2: The voice platform has launched the latest version of it’s text-to-speech model

• Claude Managed Agents: Anthropic’s agents now include “dreaming,” or a scheduled memory process 

• Google Gemma 4: Google’s suite of open models just got a speed boost with multi-token prediction

• Anthropic: Offensive Security Research Engineer, Safeguards

• OpenAI: Researcher, Trustworthy AI

• Google DeepMind: Research Scientist, Biomedical AI, DeepMind

• Cisco: AI Researcher Hybrid

GAMES

POLL RESULTS

Do you think quantum computing will be widely scalable by 2030?

Yes (64%)
No (32%)
Other (4%)