Agentic AI: Driving the Next Wave of Innovation in 2025

Recent advances in generative AI have given rise to “agentic AI” – autonomous AI systems that set their own goals and take multi-step actions. At its core, an AI agent can be defined as a combination of three components: a prompt (task instructions), an AI model, and one or more tools or actions to complete the taskaws.amazon.com. For example, AWS’s new Strands Agents framework describes an agent as simply “Model + Tools + Prompt,” with the model planning its own steps to achieve the goalaws.amazon.com. The idea is that instead of just answering a single question, an agent can plan, reason, act, and even react within a workflow. In Thomson Reuters’ words, agentic AI “goes beyond simply responding” to prompts – it plans, reasons, acts, and even reacts to complete complex, multi-step assignments with transparency and precisionthomsonreuters.com. (This emerging “agentic” paradigm underlies new AI copilot products in finance, law, and other domains.) As Thomson Reuters’ Chief Product Officer David Wong put it, “Agentic AI isn’t a marketing buzzword. It’s a new blueprint for how complex work gets done,” with AI agents operating inside professional workflows while human experts stay in the loopthomsonreuters.com.

Market Momentum and Adoption

AI agents are rapidly moving from niche demos into mainstream enterprise use. Recent surveys and deal activity show explosive growth in 2025. For example, a PwC survey of 300 US executives (April 2025) found 79% of companies already deploying AI agents in some function – and 66% of those report measurable productivity gainspwc.com. Moreover, 88% of firms plan to increase AI budgets in the year ahead because of agentic AIpwc.com, and 75% of execs agree agents will reshape the workplace more than the internet didpwc.com. In short, many companies see AI agents not as a gimmick but as a genuine competitive advantage.

VC and corporate funding underscores this momentum. Massive “megarounds” are still happening in Q1 2025: for instance, SoftBank led an $40 billion funding round for OpenAI in March (valuing OpenAI at $300B)techcrunch.com. Other AI startups scored large rounds – e.g. Nexthop AI ($110M) and Celestial AI ($250M) raised nine-figure Series rounds in Marchtechcrunch.com. Analyst and news reports agree: Silicon Valley is “all the rage” around AI agents, with VCs pouring money into companies automating sales calls, data entry, coding, and morebusinessinsider.com. For example, StackAI (a no-code agent platform) raised $16M in May 2025businessinsider.com, and recent deals include cybersecurity-agent maker Reco ($25M), workflow agent Artisan ($25M), and web-debugging agent Spur ($4.5M)businessinsider.com.

Key highlights of 2025 so far:

• Enterprise Adoption: By mid-2025, an estimated 79% of US businesses had deployed AI agents in at least some workflowspwc.com, and 88% were upping AI budgets for agentic projects.

• Funding Surge: OpenAI’s unprecedented $40B round in March 2025techcrunch.com; multiple AI startups raising $100M+ rounds (e.g. Celestial AI at $250M)techcrunch.com; many Series A/B rounds for agent-tech (StackAI $16M, Reco $25M, etc.)businessinsider.combusinessinsider.com.

• VC Interest: Business Insider notes “AI agents are shaping up to be all the rage in Silicon Valley this year”, with VCs happily funding firms that automate everything from sales calls to codingbusinessinsider.com. Industry analysts forecast 2025 as a boom year for agentic AI investments.

Key Products and Platforms

Major tech providers and startups are rolling out agentic AI products. These often embed agents into familiar apps or developer tools. For example, Microsoft declared at Build 2025 that “we’ve entered the era of AI agents.” Its Copilot tools already touch millions of users: over 230,000 organizations (including 90% of the Fortune 500) have used Microsoft 365 Copilot Studio to build custom AI agents and automationsblogs.microsoft.com. Likewise, AWS this spring released Strands Agents, an open-source SDK that lets developers quickly assemble agents by specifying a model, tools (APIs/functions), and a promptaws.amazon.com. Strands emphasizes that with today’s advanced models, developers can “simply define a prompt and a list of tools” and have agents plan and execute tasks—often going from prototype to production in days instead of monthsaws.amazon.com.

Other companies are embedding agents in industry-specific applications:

• Microsoft Copilot & Azure AI Studio: Over 15 million developers now use GitHub Copilot, and Microsoft reports “more than 230,000 organizations — including 90% of the Fortune 500” have created AI agents with Copilot Studioblogs.microsoft.com. New “agent mode” features let Copilot autonomously handle complex coding and business tasks.

• AWS Strands Agents (May 2025): An SDK allowing teams to build AI agents by combining prompts, AI models (any supporting tool-use), and a toolkit of APIs. AWS claims Strands dramatically cuts development time, letting agents plan steps and call tools without extensive custom orchestrationaws.amazon.com.

• Thomson Reuters CoCounsel (June 2025): A new agentic AI assistant for tax, audit, and accounting professionals. TR says CoCounsel “plans, reasons, acts, and even reacts” through complex workflowsthomsonreuters.com. This system uses specialized tax/audit knowledge and generative AI to automate tasks (e.g. document review, memo drafting) under strict accuracy and compliance. As TR’s VP Kevin Merlini explains, CoCounsel is not “GenAI in a prettier wrapper” but a “fully integrated, intelligent system built to do the work,” acting “with context” and following expert best-practicesthomsonreuters.com.

• Mastercard Agent Pay (April 2025): A new “agentic payments” platform integrating AI agents into commerce. Mastercard is collaborating with Microsoft and others to enable AI-driven purchasing and tokenized payments. Its press release describes “Agentic Tokens” that let conversational AI (e.g. personal shopping agents) execute secure payments on a user’s behalfmastercard.com. This initiative shows even financial services are betting on agents to handle transactions and customer interactions.

These examples illustrate how the technology is moving from research to real deployments. Companies are building and launching agentic solutions across domains— from internal workflow bots to customer-facing AI assistants. Even incumbent enterprise software firms are rearchitecting products around agents; Thomson Reuters emphasizes it has “re-architected core experiences” so that agents operate with trusted content and expert knowledge in the loopthomsonreuters.com.

Cybersecurity Implications

The surge of AI agents has both benefits and risks for cybersecurity. On the one hand, defenders see agents as powerful new tools. For example, Wiz CTO Ami Luttwak highlights that agentic AI can provide a “hive knowledge” of attacks: once one AI agent detects a threat, that intelligence can be shared globally to automatically prevent it elsewherecybersecurityasia.net. Check Point’s panel of experts similarly notes that AI agents could continuously learn to detect and neutralize known threats, hinting at a future of collaborative, real-time defense across organizationscybersecurityasia.net.

On the other hand, security experts warn of new risks from autonomous agents. Axios reports that agencies and CISOs are grappling with an “identity crisis” for AI agentsaxios.com. Unlike passive software, agents act like autonomous “non-human employees,” so firms must issue them their own credentials and privileges. As Okta’s security chief David Bradbury notes, “You can’t treat them like a human identity and think that multifactor authentication applies in the same way”axios.com. Without strict guardrails, a rogue or compromised agent could exfiltrate data or make unauthorized changes.

Industry leaders emphasize proactive safeguards. For instance, 1Password’s CEO Jeff Shiner points out that agents “work 24/7, without sleeping and at very quick speeds,” so IT teams need clear visibility into their activitiesaxios.com. Similarly, CyberArk’s security VP Kevin Bocek advises having a “kill switch” to immediately shut down any agent that “has a bad day”axios.com. In practice, security vendors are already releasing tools: just before RSA Conference 2025, 1Password and others unveiled new agent-specific identity management features, and Okta/OwnID announced products to secure AI accountsaxios.com.

Security experts summarize the situation: multiple AI agents are a double-edged sword. On defense, they promise unprecedented automation and shared threat intelligencecybersecurityasia.net. But they also require new trust models. Amazon Web Services’ Daniel Rohrer (formerly NVIDIA) stresses that fully autonomous agents will need “more observability and boundaries,” because we cannot sacrifice governance for efficiencycybersecurityasia.net. In practice, organizations must update their security practices for a world with non-human, continuously learning actors. According to Deloitte, this matters soon: they predict 25% of AI-using companies will pilot agentic systems in 2025axios.com, so the demand for agent-focused security measures is already rising.

Regulatory and Ethical Considerations

Policymakers and lawyers are taking notice of agentic AI as well. In emerging AI regulations, autonomous agents are likely to be treated as high-risk systems. For example, legal analysts note the EU’s forthcoming AI Act (a technology-neutral, risk-based framework) will cover agentic AI. Under the Act, goal-directed autonomous systems would likely fall into the “high-risk” or even “prohibited” categories depending on usekennedyslaw.com. This means companies deploying agentic AI in regulated fields (finance, healthcare, law enforcement, etc.) will face strict requirements: thorough risk management, transparency reports, and detailed logs of agent decisionskennedyslaw.com. AI products like chatbots or personal assistants are already explicitly addressed by EU rules; the analysis suggests agents (which can execute transactions and modify data) will come under at least the same scrutinykennedyslaw.com.

National guidelines echo these concerns. For instance, the UK’s recent voluntary AI Code of Practice (Jan 2025) highlights “human responsibility, auditability, [and] secure design” for autonomous AI systemskennedyslaw.com. Similarly, sectoral regulators (financial regulators, data protection authorities) have flagged “autonomy risks” from agents, indicating that official guidance on managing AI agents is forthcomingkennedyslaw.com. In practice, companies building agents will need to embed governance: implementing human oversight checkpoints, auditing agent behavior, ensuring data privacy, and documenting training data and model reasoning.

In summary, agentic AI is entering a phase of real-world scaling under growing public scrutiny. Experts caution that while these systems can drive huge productivity gains, they “must be built with real content, trained with real experts, and trusted by the professionals who do real work”thomsonreuters.com. Regulatory regimes and corporate policies will need to adapt quickly. For example, companies will likely have to maintain logs of each decision an agent makes and be able to shut agents down if something goes wrongkennedyslaw.comaxios.com.

Outlook

In just a few months, “agentic AI” has gone from theory to ubiquity in tech strategy. Large enterprises are already piloting or deploying AI agents, VCs are aggressively funding the space, and major software platforms are embedding agent capabilities by default. As Microsoft and others say, we’ve “entered the era of AI agents”blogs.microsoft.com. However, experts emphasize that success will depend on strong safeguards. In the words of Mastercard and AI leaders: agentic systems must deliver “smarter, more secure… experiences” with human oversight built inmastercard.comthomsonreuters.com. Security researchers and executives agree that building trust, control, and accountability into these autonomous systems is critical.

As one security leader predicted, 2025 will likely see the “beginnings of agents” being applied to cyber defense and enterprise workflows, with broader adoption to come in 2026 and beyondsecurityjourney.com. If handled wisely, agentic AI could empower businesses in unprecedented ways. But the consensus of recent analyses and expert statements is clear: for all its promise, agentic AI demands a commensurate focus on governance, security, and responsible deploymentaxios.comcybersecurityasia.net.

Sources: All facts and quotes above are drawn from recent industry reports and press releases (April–May 2025) from authoritative outlets and companiesthomsonreuters.compwc.comblogs.microsoft.comtechcrunch.combusinessinsider.combusinessinsider.comaws.amazon.comaxios.comaxios.comcybersecurityasia.netkennedyslaw.commastercard.com.