The Rise of AI Agents Inside the Enterprise

News

1. Tally Launches New Campaign for Saudi MSMEs Aligned with Vision 2030

2. AWS Unveils 3 AI Agents to Streamline Software Development

3. OpenAI Hits ‘Code Red’ as ChatGPT Upgrades Take Priority Over Ads

4. Apple Ousts Giannandrea, Installs New AI Chief

5. Gemini Named Most Culturally Aligned AI in UAE’s First-of-its-Kind Index

6. UAE Approves National Encryption Policy for a Post-Quantum Shift

Enterprise AI shifted gears this year as agent-style systems moved from technical demonstrations toward applied use inside companies. Models capable of planning tasks, selecting tools and taking sequential actions began appearing in operations, developer environments and consumer-facing products, advancing automation beyond last year’s generative-AI prototypes.

Recent vendor and industry-consulting reports suggest a growing interest in deploying agent-style systems across companies.

Most surveys underline the same pattern: momentum is building, but deployment remains uneven as organizations test safety controls, reliability thresholds and integration with existing systems.

Adoption is advancing in part because underlying models have improved. Better reasoning, longer context windows and stronger orchestration have widened the range of tasks agents can attempt from research assistance and scheduling to structured monitoring and multi-step automation. At the same time, these studies emphasise the limits: probabilistic behaviour, variable reliability, essential human oversight and governance frameworks still in formation.

Taken together, the findings suggest that interest in agentic AI has moved from experimentation to early planning, while broad deployment still depends on trust, accountability and the ability to integrate agents into core systems. The companies that progress fastest are those reworking workflows around these tools rather than layering them onto existing processes.

Below are some of the most visible agent-oriented platforms driving this shift. Their approaches differ, but collectively they show where enterprise automation is heading.

OpenAI Agents SDK and Responses API

OpenAI’s Agents SDK and Responses API marked a shift in how developers use the company’s models. Instead of asking a system to generate text and wait for a reply, developers can now create agents that select tools, retrieve information, carry out multi-step tasks and adapt based on the results. This moves OpenAI’s platform into operational territory, where models serve as decision layers and agents execute the follow-through within business processes. The approach has expanded the types of work that developers can automate, particularly in research, analysis, and structured integration tasks.

DeepResearch illustrates the direction vendors are exploring. In OpenAI’s demonstrations, agent workflows gather information, evaluate sources, synthesise findings and produce structured outputs with lighter supervision than traditional generative-AI interactions. These patterns hint at how higher-order reasoning tasks could be automated into pipelines. For enterprises, OpenAI positions this as a path toward agents that assist with due diligence, internal knowledge consolidation, compliance checks or competitive analysis.

Microsoft Copilot and Azure AI Agent Service

Microsoft integrated agents directly into the workflow of daily tasks across Outlook, Word, Excel, and Teams. The upgraded Microsoft 365 Copilot can schedule meetings, draft communications, summarise documents, format content, and coordinate actions across applications. Because Microsoft controls the operating system, productivity suite and cloud environment, the agent sits close to the user’s workflow and acts with a level of contextual awareness that earlier generations of automation tools could not reach. The result is a more operational assistant who participates in work.

Azure AI Agent Service opens the same capabilities to developers and enterprises that want to deploy agents at scale. The service connects agents to internal data, enterprise systems, and governance controls, providing companies with a monitored environment for automation that spans CRM tools, HR systems, logistics platforms, and custom databases.  Microsoft reports growing use of Copilot Studio to build custom agents, helped by a wide and expanding library of connectors.

For many organisations, this makes Microsoft’s ecosystem the most accessible path for early adoption.

Anthropic Claude-based Agents

Anthropic has taken a more deliberate approach to agent deployment. Claude-based agents emphasise predictability, clear reasoning and controlled actions, and are designed for environments where mistakes carry regulatory, financial or legal consequences. The company focuses on behaviours that are easier to inspect and manage, which appeals to industries that prioritise traceability. Organisations are using Claude-based agents to perform tasks that require consistency, judgement and careful handling of sensitive data rather than broad autonomy.

The positioning reflects Anthropic’s roots in safety research. The company presents Claude-based agent implementations as dependable co-workers capable of handling structured tasks within defined limits. They operate with tighter boundaries and more transparent reasoning chains than many competing systems. Organizations exploring Claude-based agents often prioritise control and auditability over automation at scale, and this approach has gained relevance in finance, healthcare, insurance and other regulated sectors.

Google DeepMind Astra and Google Antigravity

Google DeepMind’s Project Astra is an experimental multimodal agent that combines visual, audio and text inputs. Demonstrations show Astra interpreting scenes, tracking objects and responding to spoken prompts, signalling Google’s push toward agents with perceptual context. Astra remains a research prototype with limited testing access rather than a commercial product.

Google also introduced Antigravity in late 2025 as an agent-first development environment built around Gemini models. In its preview form, Antigravity gives agents access to an editor, terminal and browser so they can read and modify code, run searches and work through multi-step development tasks. The platform is still in early rollout, with features evolving and security reviews underway, but it points to a future where agents operate directly inside software environments rather than only through API calls.

Salesforce Agentforce

Salesforce describes Agentforce as the core automation layer for its CRM cloud and related services. The company benefits from a large base of structured customer-interaction data, which gives its agents a detailed understanding of sales and service workflows. Agentforce can help draft responses, update customer records, prioritise leads, route service requests, and coordinate tasks across departments. Salesforce’s control over the underlying data and workflows helps the agent operate with a level of context that many standalone tools lack.

Salesforce says Agentforce has become one of its fastest-growing products, citing adoption by thousands of customers across CRM use-cases. The company positions the platform as a core automation layer designed to streamline routine tasks while maintaining full audit trails and system-level controls. Salesforce emphasises that control of underlying data, process logic and the agent environment allows enterprises to scale automation without compromising oversight. The result is an agent model that blends autonomy with governance, appealing to regulated and process-intensive industries.

Appian Agent Studio

Appian’s Agent Studio extends the company’s foundation in workflow and process automation. Instead of treating agents as standalone assistants, Appian embeds them directly into business processes. The system lets agents work with structured data, evaluate edge cases and interact with core applications inside the same governance framework that manages existing workflows. This approach keeps automation aligned with organisational rules and reduces the risk of actions that fall outside approved boundaries.

For companies with complex, multistep operations, this design has clear appeal. Appian customers use Agent Studio to monitor processes, escalate issues, fill gaps in coordination and automate sequences that often slow down at human checkpoints. The platform provides detailed visibility into agent activity, which helps organisations satisfy compliance and audit requirements. Appian’s focus on transparency and structural alignment gives enterprises a way to scale agentic automation without redesigning their entire operational landscape at once.

Zilliz DeepSearcher

DeepSearcher, developed by Zilliz, brings agentic retrieval-augmented generation into enterprise search. The system allows agents to operate over local or controlled datasets, which is vital for organisations with strict data residency or security requirements. It uses vector search and controlled data pipelines to deliver more predictable and reproducible behaviour than broad-purpose language models. Companies can use DeepSearcher to perform internal queries, validate information and consolidate knowledge without sending data to external servers.

The platform appeals to sectors such as finance, government, defence and manufacturing, which need precise retrieval and full provenance for every answer. DeepSearcher limits agent behaviour to well-defined scopes, which reduces the risk of unexpected outputs and helps maintain accuracy. This design is well-suited for tasks such as internal documentation review, quality assurance, product support and technical analysis. It reflects a broader trend toward controllable agent systems that prioritise reliability over generalisation.