Skip to content

Let's Data ScienceLEARN • BUILD • STAY AHEAD

News
Blog
Code Problems
Pricing
Contact

© 2026 Let's Data Science

Advertise|Terms|Privacy||Image Rights

Home
News
AI Agents

Topic24 articles · Updated May 14, 2026

AI Agents News

Coverage of the fast-moving AI agent space: builders and frameworks, tool use, planning and memory, browser and desktop agents, and the enterprise platforms shipping them.

servicenow
7.1
May 14, 2026
ServiceNow Expands Autonomous Workforce Across Enterprise Functions
At ServiceNow Knowledge 2026, ServiceNow announced an expansion of its Autonomous Workforce offering, adding AI specialists for IT, CRM, employee service teams, and security and risk, according to a ServiceNow press release. The vendor presented Otto as a conversational front door and described shared governance, audit trails, and role-scoped permissions as core platform capabilities, per the press release. Forrester characterised the event as evidence of ServiceNow's ambition to become an "autonomous operating layer for the enterprise," in a Forrester blog. Reporting in TechTarget highlighted differing priorities between HR and IT when adopting agentic workflows and quoted CVS Health and HR advisory firm McLean & Co. as examples. Editorial analysis: For practitioners, the announcements consolidate a pattern where major enterprise platforms bundle agentic AI, workflow orchestration, and governance to reduce integration overhead and accelerate rollouts.
Quick Read+6
gemini-spark
6.8
May 14, 2026
Gemini Spark Leak Reveals Google AI Agent Features
NokiaPowerUser reports a leaked onboarding screen for a feature called "Gemini Spark" that appears ahead of a Google event. The leak describes Gemini Spark as an "everyday AI agent" that can help with inbox management, online tasks, connected apps, chats, websites and more, and labels the feature as Beta, per NokiaPowerUser. The leaked page lists data sources Gemini Spark may use, including connected apps, skills, chats, tasks, websites users are logged into, personal intelligence, location data and additional linked services, according to NokiaPowerUser. The onboarding screen also reportedly warns that Gemini Spark could access sensitive information and may share necessary information with third parties to complete tasks, per the leaked text cited by NokiaPowerUser.
Quick Read
microsoft-research
7.2
May 14, 2026
Microsoft Research Generates Realistic Command-Line Telemetry
According to a Microsoft Security blog post published May 12, Microsoft researchers describe an AI-driven approach that generates realistic synthetic command-line and process telemetry from attacker behaviors. The system translates tactics, techniques, and procedures expressed via the MITRE ATT&CK framework into structured logs and telemetry, and Microsoft evaluated three generation methods including prompt-engineered generation and multi-agent (agentic) workflows. Reporting summarized by GBHackers notes the pipeline produces command-line arguments, process names, and parent-child relationships intended to be semantically accurate and to exercise detection rules. Microsoft frames the work as a way to accelerate detection engineering by producing high-fidelity attack data when real-world malicious telemetry is scarce.
Quick Read
identity-security
7.1
May 14, 2026
Organizations Adopt AI Agents for Sensitive Security Tasks
According to a Semperis study published via PR Newswire on May 13, 2026, a global survey of 1,100 organisations found 74% believe AI will increase attacks on identity infrastructure. The study reports 93% already use or plan to use AI agents for sensitive security tasks such as password resets and VPN access, and 92% say AI is installed on at least some local machines with access to SSH keys and encryption keys. Globally only 32% are very confident they could regain control if AI exposes admin credentials, with 53% in the US and 12% in France. The release also reports that 65% say AI identities are fully registered, 6% do not track them at all, and among those who track AI identities 57% use the same system as human identities while 43% use separate systems. Alex Weinert, Semperis Chief Product Officer, is quoted in the release.
Quick Read+1
apple
7.2
May 14, 2026
Apple weighs allowing AI agents in the App Store
Reporting by The Information, summarized by AppleInsider, says Apple is exploring ways to allow apps that embed AI agents into the App Store while limiting safety and policy risks. AppleInsider reports The Information says Apple has so far blocked apps that can be used to write code or create programs because the App Store Review Guidelines prohibit on-device code generation and execution, and because such functionality could be used to produce malware or to create apps that bypass App Store purchases. AppleInsider also notes the company is considering technical and policy controls and that WWDC in June may include broader AI-related announcements, per AppleInsider.
Quick Read+2
identity-security
7.0
May 14, 2026
Machine Identities Outnumber Human Identities 109-to-1
Palo Alto Networks' 2026 Identity Security Landscape report, cited during the May 2026 launch of the company's Idira identity platform, finds that organizations now manage an average of 109 machine identities for every human identity. The report is based on a survey of 2,930 cybersecurity decision-makers worldwide, and projects machine identities to increase 77% over the next 12 months while human identities grow 56%, with AI agent identities expected to grow 85% in the same period. The study also reports that roughly 90% of organizations experienced at least one identity-related breach in the past year, and Australian respondents reported a 91% breach rate. Coverage of the Idira launch frames the platform as addressing machine and agent identity discovery, dynamic controls, and just-in-time privilege, and includes a direct quote from Peretz Regev of Palo Alto Networks on identity risk.
Quick Read+2
vancouver-mayor
5.1
May 13, 2026
Vancouver Mayor Clarifies Use of 11 AI Agents
At a Web Summit appearance, Vancouver Mayor Ken Sim said he has "11 AI agents" running that do much of his work, a remark that drew public concern, including from mayoral candidate Kareem Allam, who asked for disclosure on AI use, Daily Hive reports. Sim issued a written statement, quoted by Daily Hive, saying, "Any AI agents or tools I have experimented with were used strictly in a personal capacity" and that they were never run on City of Vancouver hardware or networks. Global News reports Sim described using the tools for personal learning, scanning news, tracking global and financial events, following "thought leaders," and dietary planning, and that he told Web Summit he expects AI to be "64 times better in three years."
Quick Read
tiktok
7.3
May 13, 2026
TikTok Launches Ads MCP Server for AI Agents
TikTok announced the TikTok Ads MCP server at TikTok World, its sixth annual ad product summit, allowing external AI agents to connect to its ads platform, according to Digiday and Adweek. Digiday quotes Jose Villalobos, TikTok's global head of product marketing for platform and core ads, saying the server will enable marketers to "connect their own AI agents directly to the Tiktok ads platform so their AI systems can plan, launch and optimize campaigns without manual intervention." Reporting by PYMNTS notes the Model Context Protocol, or MCP, is an open standard that moves models from passive assistants to active agents and that Google, Meta and Amazon have launched comparable MCP servers. Editorial analysis: Industry practitioners should view this as a continuation of major ad platforms exposing operational controls to agentic systems, raising data-sovereignty and auditability tradeoffs.
Quick Read+1
ai-agents
6.8
May 13, 2026
AWS and Cisco Secure AI-Agent Deployments at Scale
The AWS blog post describes a collaboration between AWS and Cisco AI Defense that addresses three enterprise security gaps for AI agents: visibility, manual security-review bottlenecks, and compliance risk. The post says MCP adoption accelerated after its November 2024 introduction and that A2A communication followed in April 2025, producing "dozens to hundreds" of MCP servers in enterprises, per the blog. The post reports risks from unvetted MCP servers, A2A agents, and Agent Skills including inadvertent access to sensitive systems, potential SOX and GDPR compliance failures, and operational disruption. The blog states manual security reviews can add several weeks to deployment. To address these gaps, the AWS blog post describes integration between Cisco AI Defense and an AWS-backed open-source project to provide automated scanning and unified governance for MCP servers, AI agents, and Agent Skills.
Quick Read
anthropic
5.6
May 13, 2026
Claude Code Creator Runs Thousands of Sub-agents Overnight
Business Insider reports that Boris Cherny, the lead engineer behind Claude Code, said during a May 4 interview at Sequoia Capital that he runs "five to 10 sessions," each containing multiple agents, and "usually, every night, I have like a few thousand" AI sub-agents performing "deeper work." Cherny told the audience he tracks those sessions on his phone using the Claude app and relies on two persistent-automation features, /loops and Routines; Business Insider reports he said /loops can be scheduled locally via cron while Routines run recurring tasks.
Quick Read
hubspot
6.9
May 13, 2026
HubSpot Announces AI Pricing Changes, Stock Falls Sharply
Reported facts: According to Search Engine Journal, on May 7, 2026 HubSpot CEO Yamini Rangan announced changes to how the company charges for its AI agents, moving toward outcome-based pricing, cutting prices for its AI customer service agents, and offering a 28-day free trial (Search Engine Journal). Search Engine Journal reports Wall Street reacted the next day: HubSpot shares closed down 19% on Friday, May 8 at $197.35, touched $180.50 intraday, are down roughly 40% year-to-date and sit about 70% below their 2021 high; William Blair downgraded the stock and Cantor Fitzgerald lowered its rating to Neutral (Search Engine Journal). The same coverage notes Q1 revenue grew 23% to $881 million, customer count rose 16% to nearly 300,000, full-year guidance was raised, the AI customer service agent resolves tickets about 70% of the time, and over 9,000 customers have activated it (Search Engine Journal). Editorial analysis: For partner agencies, this episode highlights platform coupling risk and the need to distinguish where value is delivered versus where vendor pricing or product design can change economics.
Quick Read
diffusion-of-innovations
6.6
May 13, 2026
1960s Study Foresees S-Curve AI Adoption Pattern
PYMNTS draws a parallel between Everett Rogers' 1962 study of Iowa corn farmers and contemporary consumer uptake of AI tools. Rogers documented an S-shaped adoption curve: slow initial uptake, rapid acceleration as the majority adopts, and eventual flattening; PYMNTS reports that by 1941 nearly all farmers in the studied communities had switched to hybrid seeds. PYMNTS also cites an April report estimating that AI agents could handle 15% to 25% of U.S. shopping purchases by 2030, and says current consumer use appears to be crossing from early adopters into the majority. The article frames these findings as relevant to payments and retail because broader consumer acceptance typically alters transaction flows and merchant integrations.
Quick Read
aws-workspaces
7.1
May 13, 2026
AWS Enables AI Agents to Drive WorkSpaces Desktops
The AWS blog announces that Amazon WorkSpaces can now serve as managed virtual desktops for AI agents in public preview, enabling agents to authenticate via AWS Identity and Access Management and connect to a WorkSpace using a unique pre-signed URL, per the AWS post. Agents interact with applications using screenshots (computer vision), mouse and keyboard inputs, and can store captures for audit, according to reporting by AWS, InfoQ, and heise. WorkSpaces exposes a managed MCP endpoint so agent frameworks that implement the MCP protocol, including LangChain, CrewAI, and Strands Agents, can connect, per AWS and InfoQ. Industry context: This provides a path for automation across legacy desktop applications without API modernization, but reporting also flags efficiency and cost tradeoffs for UI-driven agents versus API integrations.
Quick Read+2
gitlab
6.9
May 13, 2026
GitLab cuts jobs to fund AI agents push
GitLab announced a global restructuring that will reduce its country footprint by about 30%, flatten management layers and reorganize research and development into roughly 60 smaller teams, according to reporting by The Next Web and Business Insider. CEO Bill Staples wrote in an internal memo, quoted by Business Insider and TNW, that the changes are meant to prepare for the "agentic era" and that the company will "reinvest the vast majority of savings back into the business," per TNW. The company has not disclosed a final headcount for layoffs; outlets including Economic Times and Business Insider report the number will be determined and disclosed with quarterly results on June 2. Shares fell roughly 7-8% in after-hours trading after the announcement, per Investing.com and Business Insider. Editorial analysis: Industry observers should view this as another example of firms framing workforce reductions as reinvestment into AI capabilities rather than pure cost cuts.
Quick Read+9
ai-agents
6.6
May 12, 2026
Productive Launches 5.0 With Autonomous AI Agents
Productive announced the launch of Productive 5.0, its largest release to date, introducing AI Agents and a suite of intelligence features aimed at automating routine operational work, according to a GlobeNewswire release reported by The Manila Times. The company cites an internal study showing 76% of professionals use AI primarily for writing and editing, while only 29% use it for planning, and that "more than two thirds" of respondents would let AI handle operational tasks, per the same release. Product documentation on Productive's product updates page details Agents as custom, autonomous, and shareable virtual assistants and describes Skills as saved instruction sets that capture business procedures for reuse.
Quick Read
ai-agents
6.8
May 12, 2026
Instacart Veterans Raise $8.5M for Champ AI
Per Business Insider and Dealroom, Champ AI, founded by former Instacart engineers Jagannath Putrevu, Ted Cheng, and Peter Lin, emerged from stealth with $8.5 million in seed funding led by Redpoint Ventures, with participation from defy.vc, SV Angel, and an angel check from Instacart cofounder Max Mullen (Business Insider; Dealroom). The six-person startup builds AI agents that automate back-office tasks - logging into websites, filling spreadsheets, sending emails, making phone calls - and says it has over 10 paying customers across logistics, healthcare, and e-commerce (Dealroom). Dealroom reports Arena Club saw 30% faster card-processing after adopting Champ. Dealroom also reports the company plans to use the funding to hire software engineers and salespeople.
Quick Read
mental-models
5.4
May 12, 2026
Researchers Frame AI as Button-pushing Explorers
The Conversation published an essay on May 12, 2026, by Ji Y. Son (Professor of Psychology, California State University, Los Angeles) and Alice Xu (Ph.D. student, UCLA) that proposes a new mental model for today's AI agent systems: "button-pushing explorers," per the article. The piece notes that simple trial-and-error behaviour, take an action, observe consequences, adjust, can produce smart-seeming results without humanlike understanding. The Conversation reports that a nonprofit released a test on May 1, 2026 in which humans scored 100% while advanced AI systems scored under 1%, a disconnect the authors use to motivate the mental-model framing. The authors draw on classroom observations of students integrating AI tools to argue the model helps people recognise both capabilities and limitations.
Quick Read
ai-agents
7.1
May 12, 2026
HBR Finds Traditional Marketing Fails With AI Shopping Agents
Harvard Business Review reports that AI shopping agents are becoming a meaningful share of online shoppers and that many classic e-commerce persuasion tactics do not work on them. Per the HBR analysis, researchers tested eight common promotional mechanisms across four models, GPT-4.1-mini, GPT-5, Gemini 2.5 Pro, and Gemini 2.5 Flash Lite, using a proprietary simulation with thousands of simulated shopping rounds and found only one mechanism behaved consistently as it does for human buyers. HBR says tactics such as scarcity cues, countdown timers, strike-through pricing, and vouchers often failed or backfired. HBR also reports an exploratory survey of 50 U.S. and UK e-commerce executives that found many have noticed traffic or conversion shifts they attribute to AI agents. Editorial analysis: For practitioners, this implies A/B testing, attribution, and promotion evaluation should explicitly account for agent-driven behavior rather than assume all sessions are human.
Quick Read
ai-agents
5.6
May 12, 2026
AI coders carry half-open laptops to keep agents running
Business Insider reports that some developers are walking through airports, offices, and ice rinks with their laptops left slightly open so background AI agents can keep running. The article profiles users including Geoff Chan, identified as a 39-year-old head of product at Raven.AI, who says he leaves his laptop ajar during his daughters' skating practices to avoid interrupting OpenAI Codex-assisted work, telling Business Insider, "I have to put it up on a shelf." Business Insider reports the behavior is driven by agents that run locally or depend on WiFi, and users say shutting or sleeping the laptop interrupts progress. Editorial analysis: This pattern highlights practical operational issues for developers using long-running local agents, including battery, connectivity, and session persistence.
Quick Read
ai-agents
7.1
May 12, 2026
Researchers Propose DRIL Method to Automate Dataset Construction
A new working paper, "Deep Research on a Loop: Using AI Agents to Construct Economic Datasets" (NBER working paper W35188; SSRN abstract 6502499) proposes DRIL, a methodology that uses AI agents to assemble datasets from publicly available sources. Per the paper, DRIL applies a fixed research instrument across a mapped unit space, uses a two-stage architecture separating design from implementation, and enforces an evidence policy plus data-quality mechanisms. In a 2025 update of the Global Tax Expenditures Database for eight Latin American and Caribbean countries, the authors report the run produced 129 sources and 136 evidence records, covering 22 qualitative fields fully and 6 quantitative estimate types with documented gaps, at a cost the paper describes as comparable to a standard LLM subscription or a few hours of research-assistant work (Marginal Revolution; NBER/SSRN). Editorial analysis: This methodology demonstrates a practical agentic pipeline for routine empirical tasks, which could change how researchers budget time and verification effort when assembling primary-source datasets.
Quick Read+1
ai-agents
6.9
May 11, 2026
Article Frames Agent Design Around Operational Responsibilities
O'Reilly Radar published "From Capabilities to Responsibilities" on May 11, 2026, arguing that high-stakes, side-effectful AI agents must be designed around responsibilities rather than raw capabilities. The article describes a proposed deterministic execution kernel, called a "Kernel Space," that validates every proposed action before it touches the real world, and frames Human-in-the-Loop (HITL) as an operational bottleneck that creates governance-layer technical debt. The piece targets systems that can move money, change infrastructure, or modify critical records and presents an enforcement pattern for side-effectful systems rather than a general-purpose agent framework. The article also highlights alert fatigue and approves-throughput as failure modes when manual review scales poorly.
Quick Read
ai-agents
6.8
May 11, 2026
AI Agent Bobbi Resolves UK Police Calls
PYMNTS reports that an AI agent called Bobbi resolved 82% of citizen queries in its first week across three UK police forces without handing the call to a human. The article says the result illustrates rising adoption of agentic automation in government contact centers. PYMNTS also summarizes vendor announcements: UiPath issued a May 5 press release adding agentic capabilities to its platform for regulated agencies, enabling on-premises or self-hosted model deployments and offering data-sovereignty controls; the release included a quote from UiPath's Public Sector CTO: "As government agencies adopt agentic automation, our focus is helping them stay in control of their data, models, and how AI is used." PYMNTS additionally reports Google highlighted new tools, an open model option, and that its models now process 16 billion tokens per minute through its direct API, supported by new eighth-generation TPUs.
Quick Read
devops
6.6
May 11, 2026
DevOps Agents Balance Autonomy and Safety Decisions
According to devops.com, the article provides a framework for when an AI-powered DevOps agent should act without human approval. The piece defines a six-point autonomy spectrum from Level 0 (observe only) through Level 5 (fully autonomous) and describes four factors that determine the appropriate level for any action: reversibility, blast radius, signal quality, and time sensitivity. Per devops.com, the author recommends most teams operate agents at Levels 1-3 for the majority of use cases, reserve Level 4 for narrowly scoped actions, and consider Level 5 only after a documented track record at Level 4. The article also outlines guardrails such as approval gates, notification windows, logging, and progressive rollout to build organizational trust.
Quick Read
ai-agents
6.3
May 11, 2026
AI Agent Manages Experimental Cafe in Stockholm
An experimental Stockholm cafe is being operated behind the scenes by an AI agent named "Mona," according to reporting by AFP, Euronews and a blog post from Andon Labs. Per Andon Labs' blog, the San Francisco startup handed the lease and starting capital for a space on Norrbackagatan to Mona and says the agent handled permits, menu design, supplier sourcing and hiring during setup. Multiple outlets report that Mona conducted outreach on Indeed and LinkedIn, held phone interviews and selected at least two baristas who now work on site. Reporting by Yahoo/AFP and others says the system is powered in part by Google's Gemini. Staff and visitors have noted operational quirks, including mistaken bulk orders; barista Kajetan Grzelczak told AFP that "ordering isn't really her best suit."
Quick Read+9

Showing 24 articles

See the full news feed