Tech Talks Daily

How should finance leaders measure AI ROI when adoption has slowed and the cost of models, tokens and disconnected tools remains difficult to predict?
In this episode of Tech Talks Daily, I welcome Jeremy Ung, Chief Technology Officer at BlackLine, back to the podcast to discuss how businesses can move from finance AI experimentation to operational deployment.
Figures supplied for the interview show AI adoption in finance rising from 37% in 2023 to 58% in 2024, before moving only slightly to 59% in 2025. Jeremy argues that this apparent plateau reflects several pressures, including uncertainty around cost, regulatory requirements, auditability and the continuing debate over whether companies should build their own AI capabilities or purchase them through established platforms.
Token spending is part of the problem. Unlike traditional software costs, model usage can be difficult to predict and allocate. Finance leaders want to understand whether applying AI to a workflow will produce enough value to justify that uncertainty.
Jeremy believes companies should avoid creating artificial AI ROI metrics. The business measurements already exist. Is AI helping the company close its books faster? Is transaction matching becoming more accurate? Are collections improving? Is the work being completed faster or with fewer manual steps?
We discuss what operationalizing AI in finance looks like in practice. Many processes still require employees to contact vendors, collect information, reconcile data and coordinate with other departments. Traditional software struggled with the variation found in these workflows, while AI can adapt to different processes and communication requirements.
Accuracy and oversight remain necessary. Jeremy explains why companies need visibility into the prompts, reasoning, models, data, tools and permissions used by every AI agent. That information creates an operating record that finance teams, auditors and regulators can examine later.
His analogy with food labeling provides a useful way to understand AI auditability. Consumers can inspect ingredients, calories and sourcing information before buying food. Finance leaders should expect comparable information about the models and data involved when an agent performs financial work.
We also discuss the problem of fragmented data. Jeremy acknowledges the familiar rule of garbage in, garbage out, but argues that AI can help connect legacy platforms and mainframe systems that businesses previously found difficult to integrate.
The role of finance professionals will change as agents perform additional work. Employees may spend less time completing individual tasks and more time setting goals, reviewing results, approving actions and directing teams of agents.
Should CFOs continue buying additional AI tools, or concentrate on embedding existing investments into the financial workflows that determine business performance? Please share your thoughts with me.

What happens when an autonomous AI agent can complete thousands of actions before a traditional access review has even identified that something has gone wrong?
In this episode of Tech Talks Daily, I speak with Alex Bovee, CEO and co-founder of C1, about identity security, runtime governance, shadow AI, and the controls companies need as humans and agents begin working together.
Alex has spent much of his career in identity and security. He and his co-founder previously worked at Okta on zero trust products before creating C1 as an access control platform capable of operating at machine speed.
That requirement has become increasingly important as AI agents begin accessing company data, calling tools, using credentials, and taking actions across enterprise systems.
Alex describes agents as non-deterministic systems that can "reward-max." An agent may pursue its assigned objective so aggressively that it finds an unexpected or dangerous way to complete the task. It does not possess a moral compass or an intuitive understanding of what the company considers acceptable.
Traditional identity processes were created for people. A company might review access every 90 days or investigate a security issue after an event. That approach becomes inadequate when an agent can execute thousands of actions within minutes.
We discuss why identity is becoming a control plane for AI agents. Networks, data systems, and security tools all play important roles, but identity determines which resources an agent can access, which actions it can perform, and whether it acts independently or on behalf of a person.
Without a defined identity or delegated authorization model, an organization may struggle to connect an agent's behavior with a responsible owner, a limited mission, and enforceable permissions.
Alex explains the four connected capabilities inside C1's Agentic Control Plane.
The first concerns shadow AI discovery. Companies need visibility across cloud services, SaaS applications, endpoint agents, hosted agents, local MCP servers, and credentials stored throughout the environment.
This is particularly relevant because employees are downloading locally developed or "vibe-coded" MCP servers and running agent tools on their devices. These components can introduce software supply chain risks and expose local credentials.
The second capability covers credential security. C1 has introduced a post-quantum credential vault designed to protect secrets and inject them into authorized agent workflows without leaving credentials scattered across devices and applications.
The third area is runtime governance. Instead of reviewing behavior after an incident, organizations can evaluate an agent's actions against its assigned mission as they occur.
If an agent is authorized to complete one business task but begins exploiting an internal tool, contacting an unapproved service, or attempting to extract data, runtime controls can block the action or request human approval.
The fourth capability concerns agentic security intelligence. This uses information collected across identities, agents, permissions, credentials, and behavior to identify risks and support automated remediation.
We also discuss human accountability. Alex says emerging regulatory thinking recognizes the need for a responsible person behind an autonomous agent. That connection allows businesses to establish ownership, delegate authority, and determine who remains accountable for the agent's behavior.
The conversation then turns to the effect of AI on employees. Alex rejects the assumption that organizations will simply remove people as agents become more capable.
His preferred analogy is that people are moving from manually producing every artifact to building and supervising the factory. Employees provide the inputs, direct the agents, examine the outputs, and correct the process when necessary.
C1 has experienced this internally. Alex says its engineering team increased from roughly 150 weekly software merges to around 1,500, while engineering headcount grew by approximately 10% to 15%.
That productivity requires careful human review. Generating work faster does not remove the need to assess whether the output is accurate, secure, useful, and aligned with the original objective.
For CISOs and CIOs, the goal is to provide a governed path for AI adoption. A blanket prohibition may encourage employees to work around policy. Secure self-service access can give teams approved tools, defined permissions, and runtime protection.
If an AI agent can operate at machine speed, are your organization's identity controls capable of observing, authorizing, and stopping it at the same pace? Listen to the conversation and share your thoughts with me.

What evidence would convince you that an AI agent is ready to make decisions involving employment, money, healthcare, or legal rights?
In this episode of Tech Talks Daily, I speak with Vin Sharma, founder and CEO of Vijil, about the trust gap preventing many enterprise AI agents from progressing beyond proof of concept.
Vin has spent approximately 30 years building software across security, operating systems, open source, cloud computing, machine learning, and AI. His previous work includes leading engineering at Amazon SageMaker and helping develop 11 AWS AI services.
He argues that AI agents differ from conventional software because they combine autonomy with agency. They can interpret an objective, make decisions under ambiguous conditions, and take action. This raises a deeper question than whether an agent can complete a demonstration successfully: will it remain loyal to the interests of the person or business delegating the task?
Trust is also specific to the job. Vin uses a simple analogy. You may trust a gardener to care for your lawn, but that does not automatically make the same person suitable to babysit your child. An AI agent must therefore be evaluated within the context of its users, task, operating conditions, authority, and potential consequences.
Vin proposes testing three areas. Reliability asks whether the agent can perform its assigned task. Security examines whether it maintains its integrity when facing hostile or noisy conditions. Safety considers what happens when the agent fails and whether the resulting damage remains contained.
This evaluation cannot end when the agent enters production. Models, integrations, data, users, and external conditions change. An agent may drift away from its original purpose, which means businesses need continuous monitoring, testing, and updating across the full AI agent lifecycle.
We discuss how established security practices can be applied to this problem. Trusted execution environments, containment, least privilege, limited-duration access, and bounded models can reduce exposure. Smaller language models may also be better suited to narrow, high-risk tasks than a general model with broad permissions.
Vin offers a three-part framework for governance: personas, purpose, and policy. Personas describe the people and attackers who may interact with the agent. Purpose defines the legitimate task. Policy sets the boundaries between permitted and prohibited behavior.
For high-risk systems, his recommended starting position is that any action not explicitly permitted should be prohibited. A natural-language policy can then be converted into deterministic rules and controls governing the agent's behavior.
Vin's most direct advice concerns evidence. Vibes, demonstrations, and benchmark scores do not prove that an agent is safe for a particular business process. A CISO should expect a complete risk assessment, while a business owner should receive proof that the agent will serve the organization's interests.
His bridge analogy captures the issue perfectly. Engineers do not claim a bridge is safe because it looks impressive during a demonstration. They calculate load, tolerance, failure conditions, and provide test evidence. AI agents acting in consequential workflows deserve a comparable engineering discipline.
If an agent developer asked you to trust their system today, would they be able to provide evidence of reliability, security, safety, loyalty, and contained failure? Listen to the episode and share your thoughts with me.

What if employees could access sensitive business applications from personal phones without storing company data on those devices?
In this episode of Tech Talks Daily, I speak with Jared Shepard, CEO of Hypori, about virtual mobile infrastructure, BYOD security, employee privacy, zero trust, and the growing mobile threat created by AI.
Jared's personal story deserves attention in its own right. He describes himself as a former homeless high school dropout who joined the Army, discovered an aptitude for IT, and applied what he learned to difficult technology problems in Iraq and Afghanistan. That experience gave him a firsthand understanding of what people working at the edge need from secure communications.
The requirement that led to Hypori was unusually demanding. Users needed to obtain a phone from a local market, connect through a network assumed to be compromised, and access a protected enterprise environment without exposing sensitive information.
Hypori's answer is virtual mobile infrastructure. According to the company, applications and enterprise data remain inside a protected cloud environment while the user receives a streamed visual experience. Sensitive data is not stored on the physical phone, tablet, or laptop.
Jared explains why this differs from mobile device management. MDM attempts to secure, monitor, and control the endpoint. Hypori begins from the assumption that the endpoint may already be compromised. This can also protect employee privacy because the organization does not need visibility into the worker's personal device.
We discuss how this approach could help government, defense, healthcare, banking, and smaller businesses that cannot maintain the same mobile security resources as a large enterprise. However, virtual infrastructure does not remove every responsibility. Organizations still need strong identity controls, protected cloud environments, reliable connectivity, policy enforcement, and careful vendor assessment.
Jared also argues that AI is reducing the time between vulnerability discovery and exploitation. Security programs built around monthly patching may struggle when attack windows are measured in minutes.
The conversation closes with leadership, resilience, and mentorship. Jared explains why hard work alone does not guarantee success and why valuable lessons can come from investors, generals, colleagues, friends, or the janitor who has spent 20 years observing how an organization works.
Could virtual mobile infrastructure give employees secure access and personal privacy without forcing companies to control every device? Listen to the episode and share your thoughts.

Could your company be paying suppliers earlier than its competitors and unintentionally financing their advantage?
In this episode of Tech Talks Daily, I welcome back Oliver Belin, co-founder and CEO of Calculum. Our previous conversation took place around ten years ago when Oliver was working with the Marco Polo Network and blockchain was attracting attention across trade finance. His latest venture concentrates on working capital, payment terms, and the role of AI in supplier negotiations.
Oliver explains why working capital has moved higher on the agenda for procurement, treasury, and finance leaders. Companies can generate cash through sales, borrowing, inventory efficiency, faster customer collections, or changes to supplier payment terms. With borrowing costs higher and sales growth difficult in many markets, businesses are examining the cash already tied up within their operations.
The difficulty is that procurement teams usually know their own supplier data but lack reliable information about the terms those suppliers accept from other customers. Negotiating without market benchmarks can lead to blunt policies, such as extending every supplier to 90 days.
Oliver warns that indiscriminate extensions can create serious consequences. Smaller suppliers may experience cash flow pressure, increase their prices, reduce service, or direct capacity toward customers offering better terms. The buyer may improve its balance sheet while weakening an important part of its supply chain.
Calculum uses transactional benchmark data to compare existing payment terms with the wider market. According to Oliver, the platform can show how frequently a supplier appears in its dataset, which terms it accepts elsewhere, and the probability that it will agree to a proposed change.
AI and predictive analytics can then help companies concentrate on the suppliers where an adjustment would create the greatest financial impact and carry a higher probability of acceptance. This is particularly useful when an enterprise has tens of thousands of suppliers and procurement teams can only negotiate directly with a small proportion of them.
Oliver says Calculum typically identifies free cash flow opportunities equivalent to approximately 8% to 11% of the spend analyzed. The amount identified does not automatically become realized cash. Procurement teams need targets, internal ownership, supplier conversations, and financing options to turn recommendations into results.
He shares the example of an unnamed Fortune 500 pharmaceutical company that generated $227 million in free cash flow over 16 months. The program combined market-aligned payment terms with Supply Chain Finance, allowing participating suppliers to receive early payment in exchange for a discount based on the buyer's financial strength.
Another UK company with approximately 4,000 suppliers generated €3 million in free cash flow within two months. Oliver attributes the speed partly to knowing which suppliers to approach first rather than attempting a broad, manual campaign.
We also discuss supplier protection. Calculum identifies whether a business is a small or medium-sized enterprise, examines ultimate ownership, and considers financial strength. A financially vulnerable supplier may need early payment support rather than longer terms.
Oliver's wider point is that AI cannot create reliable benchmarks from nothing. Useful predictions require traceable transactional data, clear objectives, and people prepared to act. Could better payment term intelligence improve your cash position while creating fairer, better-informed supplier relationships? Listen to the episode and share your thoughts with me.