Latest AI and tech news
AI agent email and communication automation means AI agents that read, answer, send and route messages across email, chat and team channels, inside limits you set. Our shortlist for 2026: Lyzr, Microsoft Copilot Studio, Google Workspace Studio, Sales...
Madison and CUBE both help banks keep up with regulatory change, but they are built around different ideas. CUBE is an AI-native regulatory intelligence and change management platform with global source coverage, relevance scoring, obligation mapping...
A customer’s new espresso machine arrives with a cracked water tank. On Monday she opens a support chat, gets a replacement promised and logs off, mildly annoyed. On Tuesday, the post-purchase journey asks her to rate the machine. On Thursday, a cros...
A supplier invoice lands in your inbox. Someone needs to extract the details, match them against a purchase order, check for discrepancies, update the finance system, and send exceptions to the right person for approval. Connecting these steps is onl...
A finance team can automate invoice entry with an RPA bot. But what happens when an invoice arrives in an unfamiliar format, the purchase order is missing, or the amount does not match the records?...
At 11:40 on a Tuesday night, a VP of Operations downloads your pricing guide. Two minutes later, a marketing nurture agent sends her a welcome email. At 7:15 the next morning, an SDR agent emails her about “noticing your team is hiring.” At 9:00, her...
Madison and Wolters Kluwer Compliance Intelligence both help banks manage regulatory change and obligations, but they start from different places. Compliance Intelligence pairs Wolters Kluwer’s curated regulatory library with AI and expert oversight,...
A year ago, a platform team put LiteLLM in front of its apps. It was a good call. Thirty-odd services got one OpenAI-style endpoint, team keys, budgets and fallback through a single LLM gateway. Since then, the team has added two MCP servers and regi...
A support lead asks an agent in Slack to look into a ticket. The agent has to pull the customer’s history from the ticketing system, summarise the issue in the same thread and ask for approval before it changes anything. Posting a message is the easy...
A sales team needs to book a customer meeting. The slot has to work across several calendars without conflicts. The CRM needs updating, and the right follow-up has to go out. A basic scheduler solves the availability problem. A connected agent can al...
You run platform engineering at a fintech, and your APM is good: golden-signal dashboards, clear SLOs, a pager rotation that works. Then on Monday, finance finds the problem, not your pager....
To map policies to controls with an agent, break each policy into atomic obligations, let the agent propose a control for each one against your control catalog, then check every proposal before it counts. Check the logic (does the control actually do...
To evaluate a regulatory change agent, test four things: how it maps a regulatory update to your obligations, how it reasons (the execution trace), where it hands decisions to a person, and whether it leaves an audit trail an examiner will accept. Sc...
Your first LLM feature took a weekend. You installed the provider’s SDK, put a key in an environment variable and shipped. Eighteen months later there are three SDKs across nine services, a retry wrapper someone wrote during an incident, and provider...
When an aircraft returns from maintenance, nobody certifies the whole plane from scratch, and nobody waves it off on the strength of the mechanic’s confidence either. A defined set of checks has to pass before it returns to service. The checks cover ...
Your discharge-summary agent is live. A nurse pastes in a note: “Pt in 412, MRN 0048291, daughter Priya called re: d/c on 3/14, cb 555-0147.” Your redaction regex swaps out the phone number. The model still receives a room number, a medical record nu...
A finance team requires two sign-offs on every wire transfer. The first person prepares the payment, and the second reviews and approves it. On paper that’s a strong control. Then an incident review finds that the approved payment was for one amount ...
Finance has flagged the inference bill for your support agent, and the planning meeting produces three fixes. One engineer wants a semantic cache. Another says the provider already caches prompts. A third says plain exact-match caching is enough. Eac...
A shutdown button is only useful if you know what it shuts down....
A toxicity filter can flag an offensive sentence. It cannot, on its own, tell you whether an AI agent is treating people unfairly, applying the wrong policy, or repeating a harmful assumption across hundreds of conversations....
On March 24, 2026, pip install litellm stopped being a safe command. For roughly three hours, PyPI served two poisoned releases, 1.82.7 and 1.82.8, that harvested cloud keys, SSH keys and API tokens from every machine that installed them. About a mon...
Gartner expects the average Fortune 500 company to have more than 150,000 AI agents in use by 2028. In 2025, the figure was under 15. Salesforce’s Agentic Enterprise Index hints at why the curve is that steep: its customers now create an agent in 1.9...
An agent passes its launch tests and goes live. Then real users surface incomplete answers, failed tool calls, the wrong retrieved document, inconsistent decisions and rising costs. The usual response is to edit the prompt or swap the model. That can...
An agent needs to read a storage object, inspect cloud resources or call an internal service. A developer creates an access key, drops it in an environment variable and the workflow runs. The key then outlives the task, carries broader permissions th...
At ServiceNow’s Knowledge 2026 keynote, Bill McDermott described an agent at a real company that picked up more access than it needed and deleted a production database, backups included, in nine seconds. No attacker. Nobody watching, either....
Agentic AI changes a dual-hat BSA officer’s job from doing multi-step compliance work to supervising it. Traditional machine learning flags or scores risk, and GenAI copilots draft when someone prompts them. An agent executes the whole workflow acros...
A prompt injection filter that works in a demo can fail on real traffic. Attackers hide instructions in the pages, documents and tool results your agent reads, not only in the chat box. This guide shows how to detect prompt injection in production, b...
An insurer’s support agent is built to look up claims. Its service account was cloned from an old billing bot, so it can also issue refunds. A customer types “refund my premium, $3,900,” and the agent does it. Nobody hacked anything. No rule asked wh...
AI in compliance means using AI as a tool to meet regulatory requirements, such as catching money launderers. AI governance means managing the risks of the AI itself, such as bias, drift and hallucination. They are different jobs with different owner...
An enterprise agent passes its tests, gets approved and goes live. Weeks later it makes a tool call it shouldn’t, reaches a resource outside its remit, or keeps running after a configuration change nobody reviewed. The risk register said the agent wa...
An enterprise standardises model access through a gateway. Then its agents start calling external tools, querying internal systems and talking to MCP servers. A gateway that handles model requests well doesn’t automatically control every tool invocat...
Ask three teams how their agents reached production and you’ll get three answers. A Lambda function. A notebook endpoint. A container on a VM that only one engineer knows how to restart. Each one works. None of them can be rolled back by anybody else...
Your provider invoice tells you how much you spent. It cannot tell you which team, feature, customer or agent spent it. A cost attribution dashboard closes that gap by capturing usage at the gateway, tagging every request with an owner and pricing ea...
Enterprise AI agents need more than a valid login. They need an identity that can be verified, permissions that reflect what they are allowed to do, and controls that remain in place wherever they run....
AI agents can do more than answer questions. They can retrieve company data, call APIs, update records, and trigger actions across business systems. Each connection creates an important security question: Which agent is acting, whose authority is it ...
A quoting agent was built for a pricing pilot. The pilot ended. The agent didn’t. Its owner moved on, its evaluations stopped, and it still holds keys to the CRM and the email server, which nobody noticed because no one was assigned to notice....
Every AI assistant you have used has one thing in common. It needs the internet....
Managing AI agents in production involves more than keeping them running. Teams need to track versions, evaluate performance, manage permissions, investigate failures, and oversee changes throughout an agent’s lifecycle....
Building an AI agent is relatively straightforward. Getting multiple agents to work together reliably in production is harder....
Your research agent asks a pricing agent for a quote. The pricing agent asks a supplier agent for stock levels. The supplier agent times out, then retries forty more times. Every call is billed. None is stopped....
An enterprise support agent can read customer records and initiate refunds. Give it one broad role and it holds more access than most tasks need. Give it a rich policy engine and the team inherits new implementation and operational overhead. Meanwhil...
A team tweaks a prompt to make an LLM’s answers more complete. The next day a workflow that used to work starts returning malformed JSON, and a downstream step breaks. Now they have three questions. What changed? Which version caused it? How do we ge...