Skip to main content
thinking

Navigating the AI agent landscape

by John Ennew

10/9/2025

AI Agents Blog Image (2)
A head with the words AI in centre surrounded by symbols

We consider the role AI agents will have in the future world of work within an organisation and some of the benefits and risks for an organisation.

What is an AI Agent?

The idea of automating processes is as old as the human race and constant improvement to process automation is the story of our technological evolution.  AI Agents are just the next step in a chain that extends from the wheel, to the printing press, to the computer and to the internet.

Where previous technologies still required a level of human operation, the idea of an agent is it has agency.  By this we mean it has the ability to reason about what it is being asked to do and so can be given high level goals to achieve and work out what it needs to do rather than being directed in its every action.

AI Agents have been made possible by the invention of the large language model (such as Gemini and GPT5) which can synthesise your data alongside human language, human speech and images to deliver novel outputs not previously possible by automated workflow tools.  

You are likely already using AI Agents which are built into existing tools.  If you’re making use of the deep research capabilities of some of the popular AI tools then you’re making use of AI Agents.

Are today’s Agents truly autonomous?

The short answer is no. Any Agent we build will still be bounded in its scope and operation and directed by humans. And this is a good thing. At today’s level of Artificial Intelligence, we still need humans to be accountable for what these systems do.

This does not stop agents developed today being extremely useful tools in the workplace.  

An AI Agent differs from a traditional software application in the ease at which they can be constructed and the purposes to which they can be applied.

What advantages can Agents bring to my organisation?

Agents can be used to:

  • Provide powerful workflow tools to help members of your organisation achieve their goals with less effort. 

  • Increase the throughput of work through departments without the need to scale the department.  

  • Uplift quality in some situations.  Because work can be completed quicker, more time can then be spent reviewing and refining the outputs. 

  • Be a powerful onboarding and training tool, giving new members of staff the ability to get up to speed quicker and answer questions and give recommendations as they work.  

  • By automating repetitive and time-consuming tasks, organisations can reduce operational costs associated with manual labor. This can free up human employees to focus on more complex, strategic, and creative work.

  • Enhanced decision-making: By processing and analysing large datasets quickly, AI Agents can provide insights and recommendations that support better, faster, and more informed decision-making. They can identify patterns and trends that might be missed by human analysis.

A recent study from Harvard Business school  backs up these observations. The study reviewed the opportunities from Generative AI on an organisation.  They gave the same task to two groups. One group didn't have access to support from an AI Agent and another did.  The results showed improvements in productivity for all levels of knowledge workers, but, interestingly, the biggest boost was  noted for new employees and less experienced people.

The study’s key findings were:

  • 40% improvement in quality

  • 25% of tasks completed faster

  • 43% average performance improvements

  • 12% more tasks completed

What are the risks?

Another recent study in Nature  looked into some of the long term risks of implementing AI Agent workflows in an organisation on the staff using these tools.

This study highlighted some of the long term psychological costs of using this new technology.  While productivity in this study once again showed improvements in overall productivity inline with previous research, it also showed that workers' intrinsic motivation with their work in some areas was affected.  

Where AI Agents are stepping into the most rewarding aspects of a job, the human’s innate needs of autonomy and competence is compromised, leading to boredom and lack of curiosity, key requirements for growth and personal development.  

It’s important for leaders in organisations to consider these factors and design any AI Agent workflows in a way that allows a level of collaboration between humans and AI systems.  It is also important to understand that we are likely to still require a human expert to be accountable for anything the AI produces.

There’s been some recent research which has looked at the success of AI introduction projects into organisations. A recent report from MIT  found that 95% of AI pilot projects are currently failing.  Some of this is attributed to businesses getting to grips with the new technology and learning to understand where it can be applied successfully, they are pilot projects after all so a certain amount of failure is to be expected. There were still concerns found about bias issues and hallucinations where AI Agents are found to produce the wrong results in some contexts.

But another key reason for the failure has been attributed to the human transformation and change factor. If the people using the tools aren't bought into the advantages and taken along the journey, as with any digital transformation project, the results can be disappointing.  

The best practice advice for any change project would therefore be relevant when rolling out AI systems:

  • Ensure sponsorship from the executive level and embed the changes in your strategy.  

  • Develop good communications and set up working groups or AI councils to connect representatives across the organisation who can help champion and navigate the change. 

  • Run pilots and report back to the council often to demonstrate progress as well as assessing risks.

What Agents could I build?

At manifesto, we have conversations with clients about how AI can fit into their workflows and support them in identifying the sorts of Agents that could be  developed. 

Identifying good candidates for potential AI Agent support does need careful consideration, taking into account the organisation, the people and the tasks. As mentioned in the risks above, the biggest challenge is often bringing the people along the journey and working with team members to understand what’s possible and what the opportunities will be for them.

Starting with smaller, well-defined problems often leads to more successful pilot projects.

With that in approach in mind, after mapping out existing workflows, we would look for potential candidates based on some of the following criteria:

  • Repetitive tasks: Tasks that are done frequently and consume a lot of human time.

  • Information retrieval challenges: Situations where employees spend too much time searching for answers.

  • Areas needing quality uplift: Where more time for review and refinement could significantly improve output.

  • Onboarding and training needs: Areas where new staff struggle to get up to speed quickly.

  • High-volume, low-complexity interactions: Ideal for initial automation to free up human staff.

Some examples of AI Agents to consider:

  • Internal support agent.  Provides a natural language interface to your knowledge base to more quickly find answers to queries staff might have. This could be surfacing policy or HR related queries which are normally directed to an operations team member.

  • Content Creation / Marketing Assistant. Generates drafts for marketing copy (social media posts, email newsletters), blog outlines, product descriptions, or even internal communications. Can also help brainstorm ideas or rephrase existing content.  This can speed up content production, help overcome writer's block and assist in maintaining brand voice consistency.

  • Data dashboard. Pull together data from several spreadsheets or other internal systems to quickly present key performance indicators (KPIs) or project progress, offering a consolidated view for stakeholders without manual aggregation.

  • Research assistant. Gather and synthesise information from internal databases and documents on given topics.  We’ve also set up news summaries which aggregate and summarise the latest news feeds on given subjects so that staff get a highlighted report for the latest developments in the industry on their focus areas. These can also recommend press releases the organisation may want to consider developing in response to situations happening in the world.

How can I build my own Agent?

From a technology implementation perspective we’re classified the tools into four levels:

  • Level 1: Conversational AI Platforms / No-Code Agent Builders. These platforms enable users with minimal or no coding experience to create simple, task-specific AI agents primarily through natural language prompts or guided interfaces. They are ideal for automating straightforward request-response interactions and can be thought of as "AI assistants" for specific, contained tasks.  Examples at this level include Google’s Gemini Gems, or OpenAI’s ChatGPT (Custom GPTs).

  • Level 2: Low-Code/Visual Development Platforms. These tools offer visual development environments and pre-built components, allowing users to create more sophisticated applications and workflows with limited coding. They facilitate the integration of data from various sources and are well-suited for building interactive dashboards, data processing applications, and more structured agent behaviors.  They are able to handle more complex logic and data manipulation compared to basic conversational agents.  Google's AI Studio is a good example of a tool at this level.

  • Level 3: Workflow Automation & Integration Platforms. These platforms specialise in orchestrating complex, multi-step workflows by connecting various applications and services. They provide extensive libraries of connectors and visual builders to define intricate process flows, making them suitable for integrating AI agents into existing enterprise systems and automating cross-application tasks.  Examples at this level include n8n, Zapier or FlowWise. These tools are a step up from the previous level as they specialise in connecting disparate systems and enabling end-to-end automation, often involving multiple AI and non-AI components.

  • Level 4: AI Agent Development Frameworks / SDKs.  These are open-source software development kits (SDKs) and frameworks designed for experienced developers to build highly customised, robust, and scalable AI agent systems. They offer granular control over agent architecture, memory, tool integration, and orchestration, enabling the creation of advanced multi-agent systems and complex reasoning capabilities.  Examples at this level include LangChain, AutoGen from Microsoft and Google’s Agent Development Kit (ADK).

Organisations we’ve worked with can get started at Level 1 almost immediately after some basic introductory training.  Level 2 often needs some experimentation and support to be used internally but can be harnessed by the organisation itself within a month or so after training. Level 3 and 4 tools will typically require support either from external consultants or by building an internal capability, typically within the IT department.  

As these systems are connecting and processing your organisation's data, a review of any tool and its data policies will need to be undertaken to ensure it is compliant with your rules and policies.

In conclusion 

AI Agents represent the next significant step in automation, moving beyond traditional software to systems with genuine agency that can reason and achieve high-level goals. 

These agents can offer substantial benefits to organisations, including enhanced workflow efficiency, increased productivity, improved quality, and powerful onboarding tools. 

For anyone starting out on their AI journey and would like support in understanding the technology, risks and opportunities then please do get in touch.