Introduction
Welcome to another episode of Digital 4 Productivity, the TJ Technology Tuesday and the AI Booster. Today we have an episode on the topic of agents.
Yes, agents — not in the sense of James Bond, but in the sense of: what kinds of agents are currently available on the market? What is the idea behind them? I have put together an overview for you and we will take a look at it together.
AI Agents at a Glance
AI Agents 2026 — that is the idea, and by the way, it was created with Claude, so with AI assistance as well. Let us take a look at this together and dive into the presentation, into the overview I have prepared for you.
We have the topic of enterprise platforms, we have browser agents, workflow automation, and personal multi-agents. Let us go through what is available in each of these four areas.
We start with enterprise platforms. The best-known: Microsoft 365, the environment whose rollout I support companies with. Then of course its counterpart, Google Gemini Enterprise. And then there are naturally the OpenAI and Anthropic business solutions, as well as Agentforce by Salesforce in the CRM space.
Microsoft's AI Agent
If we look at Microsoft 365 Copilot: if you would like the slides, simply leave a comment below and I will gladly send them to you. The great strength here is that you have deep integration into the entire Microsoft 365 universe. In other words, it practically bathes in your content, and the big advantage is that everything is hosted in Germany. That means you are on the safe side, because your data resides in Frankfurt, Germany — or rather, the AI processing is carried out within the EU.
Gemini Enterprise
The counterpart, Gemini Enterprise — I have marked it as conditionally suitable here: however, with Google Gemini you also have the option, for the appropriate fee, to have everything hosted on servers in Frankfurt. It is essentially the alternative to Microsoft 365.
Google's AI Agent
Google essentially offers three models: you pay with your data, you pay with advertising, or you pay with money. And if you pay with money, you can go all the way up to the enterprise version. Airbus, for example, uses Google services — and as a defence company, you can imagine they are very security-conscious.
ChatGPT Enterprise
Then there is ChatGPT Enterprise — here too you have the option, just as with Claude, to use these services in an enterprise version. So again, a range of options, all the way to local installations. With ChatGPT I always have the question: okay, I am on American servers and I am dealing with Sam Altman. That means the US corporation issue arises here too — and also the question of ethical alignment. From my perspective, Claude — meaning Anthropic as a company — is set up somewhat more cleanly in that regard.
The Partnership Between Anthropic and Microsoft
In terms of performance, I have to be honest: Claude Cowork is extremely capable and has the great advantage of offering a very polished user interface — it works as its own standalone application. And what is interesting — you can see it on this slide as well — is the Microsoft partnership. Microsoft has entered into a partnership with Anthropic, meaning that from May onwards, you can use Cowork directly within Microsoft 365. And in combination with Purview, the compliance component, you can also configure everything cleanly from a compliance standpoint.
The idea behind Microsoft is that they are not trying to do what Apple does — attempting to get their own models to work and then failing. Instead, Microsoft is taking a cooperative approach: they have a partnership with OpenAI, currently using the 5.5 model, and they also have a partnership with Anthropic, integrating Claude Cowork. A genuinely interesting approach — being able to integrate the best models into the Microsoft 365 universe.
From my perspective, that is also what is most interesting for companies. Claude Cowork as a standalone is only conditionally suitable — but when used in combination with Microsoft 365, it becomes a genuinely compelling proposition.
Comet — The AI-Powered Browser by Perplexity
Next, let us look at browser agents. From Perplexity there is the Comet browser. It has had a few security issues in the past. However, for topics that are not security-critical, it is currently the most productive browser in my view. You may already know from Microsoft Edge that there is a sidebar where it can theoretically do things.
What Edge can do in theory, Comet can do in practice. For example, if you are building flows in Microsoft Power Automate, Comet can support you exceptionally well — it can research topics. It is genuinely the most capable option by a wide margin.
There is also a browser agent from ChatGPT. Here, again, I have to be honest — the Sam Altman question arises once more. But for research purposes, it is also worth a try.
Workflow Automation
Now let us look at workflow automation. My number one here: Power Automate. And what is quite interesting: many people tend to say, yes, Power Automate — that is somehow boring, unsexy. But the interesting thing is that it offers the deepest M365 integration of any tool — I will walk through the strengths in detail. When you work with Microsoft 365, my experience from projects bears this out: I am always called in and told to build some AI, and then it turns out to be 80% Power Automate topics and 20% AI.
You also have an AI Builder here with GPT models. You can — we did a project, for example, reading PDF documents directly and similar tasks. You have the compliance angle, which you can address wonderfully. And it is already included in many licences.
The downside: Power Automate is not entirely self-explanatory — you do need to invest some time into it. The good news is that you can get support from Perplexity.
There are also connector limits outside the Microsoft world. The good news is that you can add connections to other systems — it just costs a bit extra, but you can do it. That means it can sometimes get a bit more expensive, and debugging can occasionally be a little cumbersome. What is interesting is that you can really do things that you do within the Office environment. There is also Power Automate for Desktop as an RPA component — Robotic Process Automation. That is a separate licence and works somewhat differently, but is also interesting. The key point is that your data stays on Microsoft servers in Frankfurt — which means you are GDPR-compliant. Always with the caveat: yes, we are still talking about a US company.
Automation with n8n
n8n is a Berlin-based company, and with n8n you have the option to either host with them or alternatively run everything on your own servers, or on Hetzner or Hostinger — German servers. The idea is that, similar to Power Automate, you have triggers and then workflow steps that you can combine in deterministic and AI-driven structures.
The advantage of n8n is: if you do not have Microsoft 365, you get the automation component here. Advantage number two: you can even host it in-house. Advantage number three: there are a very large number of connectors available. Now, there are also enterprise licences, which means it does not stay quite as affordable. And the learning curve is relatively steep, relatively complex. And honestly, I have to some extent skipped over n8n for myself, because I move straight into the topic of Vibe Coding with Loveable, where instead of building a workflow I pour it all directly into a piece of software. That is always my additional consideration. But n8n is still a German company — Aleph Alpha has just been acquired by a Canadian company, but n8n remains a German company and is extremely flexible.
Automation with Make.com (formerly Integromat)
In this space, the old Integromat is now Make.com, based in Prague. The parent company is Celonis in Munich. That means we are within the EU and we have a German company as the parent. The advantage here is a significantly lower learning curve. You always have a trigger and an action and say: when this happens, that happens. You have a visual, intuitive builder. The learning curve is far lower than with other tools.
The technical foundation is the oldest relative to n8n and other tools, but it is genuinely very good as a starting point. You can really click things together with drag and drop.
Automation with Zapier
The classic — I have included it here once more: Zapier is a data hub. Their business model is that they have thousands of applications that have simply implemented a Zapier interface.
And every application that has a Zapier interface can then be used to say, for example: I have a contact who signs up for a newsletter, and they are then automatically added to my CRM system — even though both systems cannot directly communicate with each other. It is truly a fantastic tool, rock solid — I have been using it for 15 years to keep my newsletter and CRM in sync.
Multi-Agents: OpenClaw, MaltBot
Then there are personal multi-agents. Some of you have certainly heard of OpenClaw, MaltBot, or similar tools — now part of the OpenAI team. You can already see from the warnings listed: Hot warning — do not use on production systems. The idea is that this agent is practically self-learning and does impressive things on its own. I would keep it as far away as possible from any live systems, especially email systems.
You do need to be careful here — but for marketing purposes it is already wonderful and a compelling, interesting approach.
AI Agent: Manus
Then there is Manus, now part of Meta — which is a red flag for me personally. It was previously based in Singapore, highly productive, truly impressive. Though by now I would say I actually find Claude Cowork even a bit better. And many things I used to do with Manus I now do through the Claude plugin integration in Excel. There is a plugin that I can place in the sidebar and use to research things directly in Excel.
The same is available in PowerPoint. So I have to say, Claude has become my go-to — it is really outstanding.
Hermes
OpenClaw was once the big thing. Now many say Hermes is the new big thing. It is essentially a competing product to OpenClaw, where you do need a certain level of technical affinity. The advantage is that, like OpenClaw, it can be installed on your own local machine or your own server, or hosted at Hetzner or Hostinger.
So here, perhaps, a brief overview: Enterprise, Browser, Workflow, Personal. For enterprise, my clear recommendation is Microsoft 365 Copilot, with Power Automate as the automation tool where I can genuinely accomplish a great deal. For those working with Google: Gemini Enterprise. There are also options with ChatGPT and Claude Cowork.
For browsers: Comet, currently the clear leader in my view and significantly more capable than what Microsoft offers in the Edge browser. For workflows: n8n and Make.com as EU options, plus Zapier — with n8n having the steepest learning curve, while Make.com and Zapier are considerably more accessible. Both Make.com and Zapier are increasingly adding features that were previously only found in n8n. And then there are what I would call personal multi-agents — with tools like Paperclip, for instance, where I can combine different agents with one another.
The Agent-to-Agent Standard
One more very interesting development to close: there is a new standard — the A2A Agent-to-Agent Protocol. Back in November there was what you might call the first revolution: the MCP protocol. MCP — the idea being that different systems can communicate with one another. And now it goes a step further. The Model Context Protocol, or MCP, was developed by Anthropic.
Last November it was donated to the Linux Foundation, making it open source and adopted by other companies as well. Virtually everyone supports this protocol. And now it goes one step further: A2A Agent-to-Agent, enabling agents to work together with one another. Several players are already on board. It is not quite as mature yet — it has only existed since April 2026, so it is very, very fresh — and more will be joining.
Conclusion
Sources, as always, are available on request. If you would like the slides, simply leave a comment below and I will gladly send them to you. If you need support in deciding which agentic systems make sense for your organisation and which do not, please feel free to get in touch. I am happy to help.
Yours, Thorsten Jekel.
Key Takeaways
- In 2026, AI agents fall into four main categories: enterprise platforms, browser agents, workflow automation, and personal multi-agents.
- Microsoft 365 Copilot is the most recommended enterprise platform, as data is hosted in Frankfurt in GDPR-compliant fashion and it offers deep integration into the M365 universe.
- Microsoft partners with both OpenAI (GPT-5.5) and Anthropic (Claude), meaning Claude will be available directly within Microsoft 365 from May 2026 onwards.
- The Comet browser agent by Perplexity is currently the most capable AI-powered browser and surpasses in practice what Microsoft Edge promises in theory.
- Power Automate is the leading workflow automation tool for Microsoft 365 environments; in practice, many AI projects turn out to be 80% Power Automate tasks.
- n8n (a Berlin-based company) offers maximum flexibility and allows self-hosting on your own servers or German providers such as Hetzner or Hostinger.
- Make.com (parent company Celonis, Munich) has a significantly lower learning curve than n8n and, as an EU-based solution, is an excellent starting point for automation.
- Zapier serves as a proven data hub with thousands of app connectors and has been used productively by the presenter for 15 years.
- Personal multi-agents such as OpenClaw, MaltBot, or Hermes should not be used on production systems — particularly not on email systems.
- The new A2A protocol (Agent-to-Agent, launched April 2026) builds on the MCP protocol developed by Anthropic — now open source — and enables agents to collaborate directly with one another.
Frequently Asked Questions
What are the four categories of AI agents in 2026?
In 2026, AI agents fall into four categories: enterprise platforms (e.g. Microsoft 365 Copilot, Google Gemini Enterprise, ChatGPT Enterprise, Claude), browser agents (e.g. Comet by Perplexity), workflow automation tools (e.g. Power Automate, n8n, Make.com, Zapier), and personal multi-agents (e.g. OpenClaw, MaltBot, Manus, Hermes).
Why is Microsoft 365 Copilot recommended as the best enterprise solution?
Microsoft 365 Copilot is recommended because it is deeply integrated into the entire Microsoft 365 universe and AI processing takes place in GDPR-compliant fashion in Frankfurt within the EU. This means companies are on safe ground when it comes to data protection and compliance.
What is the connection between Microsoft and Anthropic (Claude)?
Microsoft has entered into a partnership with Anthropic, meaning Claude (specifically Claude Cowork) can be used directly within Microsoft 365 from May 2026 onwards. In combination with the compliance component Microsoft Purview, the solution can also be configured cleanly from a data protection standpoint.
What is the Comet browser by Perplexity and what is it used for?
Comet is an AI-powered browser by Perplexity, currently regarded as the most capable browser agent on the market. It is particularly well-suited for research and for supporting tasks such as building Microsoft Power Automate flows — however, it should only be used for non-security-critical content, as it has had security issues in the past.
What are the key strengths of Power Automate as a workflow tool?
Power Automate offers the deepest integration with Microsoft 365, an AI Builder with GPT models, GDPR-compliant servers in Frankfurt, and is already included in many Microsoft licences. In practice, automation tasks with Power Automate make up around 80% of the effort in typical AI projects.
What distinguishes n8n from Make.com and Zapier?
n8n (from Berlin) offers maximum flexibility and self-hosting options but has the steepest learning curve. Make.com (EU-based, parent company Celonis in Munich) is significantly more beginner-friendly and uses a visual drag-and-drop approach. Zapier is the oldest and best-known data hub, with thousands of connectors and is also easy to use.
Who is n8n particularly suited for?
n8n is particularly suited for users without Microsoft 365 who are looking for a powerful automation solution and are willing to invest in a steeper learning curve. An additional advantage is the ability to host n8n in-house or on German servers such as Hetzner or Hostinger.
Why is caution advised when using multi-agents such as OpenClaw or Manus?
Personal multi-agents such as OpenClaw or MaltBot should explicitly not be used on production systems — particularly not on email systems. Manus now belongs to Meta, which raises data protection concerns; however, for marketing purposes such agents are certainly interesting.
What is the MCP protocol and why does it matter?
The Model Context Protocol (MCP) was developed by Anthropic and enables different systems and AI models to communicate with one another. In November 2025, MCP was donated to the Linux Foundation, making it open source and now supported by virtually all relevant companies in the AI industry.
What is the A2A protocol and how does it differ from MCP?
The A2A protocol (Agent-to-Agent) is a new standard introduced in April 2026 that builds on the MCP protocol and enables direct communication and collaboration between different AI agents. While MCP establishes the connection between systems and models, A2A goes one step further and allows agents to cooperate independently with one another.
Tools & Resources Mentioned
- Microsoft 365 Copilot — Enterprise AI agent with deep M365 integration and GDPR-compliant hosting in Frankfurt
- Claude (Anthropic) — AI model by Anthropic, available as Claude Cowork within Microsoft 365 from May 2026
- Microsoft Power Automate — Workflow automation tool with deep M365 integration and AI Builder
- Comet Browser (Perplexity) — AI-powered browser agent, currently the most capable on the market
- n8n — Open-source automation platform from Berlin with self-hosting option
- Make.com — EU-based automation platform (parent company Celonis, Munich) with a low barrier to entry
- Zapier — Proven data hub with thousands of app connectors
- Google Gemini Enterprise — Enterprise AI agent by Google, optionally hosted in Frankfurt
- Agentforce (Salesforce) — AI agent platform in the CRM space
- Hermes — Alternative personal multi-agent, self-hostable on your own servers or at Hetzner/Hostinger
