Collaborative Agentic Workflows Keep Users In A Productive Flow State

Today, we stand at an exciting intersection of design, technology, and artificial intelligence. As more of us turn to AI to enhance our work—whether it’s designing new products, developing complex systems, or creating content—we’re starting to see both the potential and the limits of these tools. One of the most powerful trends in AI today is the emergence of agentic workflows, which might just be the key to making AI agents a valuable part of our collaborative teams.

The Current Problem: Fragmented AI Tools Lead to Fragmented Ideas

Let’s be honest: when we use AI today, it often feels like a patchwork of different tools that don’t quite fit together. For example, if you’re writing a well-researched article or whitepaper, you’ll probably use one AI tool to research, like Perplexity for searching and summarizing data, and then another AI, like Claude, to write the content.

But here’s the problem: those two tools don’t talk to each other. The insights you gained during the research phase often get lost in the transition to the writing phase. This hand-off process—usually involving copy-pasting—disrupts your flow as the user and degrades the quality of the final output.

In short, context gets lost, and you end up spending more time reorienting yourself than actually creating. It’s the digital equivalent of a relay race where the runners don’t pass the baton smoothly—and the result is a drop in speed and precision. Apologies to my track coach for all those times I fumbled the changeover of the baton!

Why Should These Tools Work Together?

Just like in human collaboration, when AI agents are working toward a common goal, it’s important that they share the same vision and understand each other’s roles. Right now, tools like Perplexity and Claude act independently, focusing only on their individual tasks. Perplexity excels at searching for relevant information and summarizing it with citations. Claude shines at writing cohesive, structured content. But neither knows that the final objective is a cohesive, well-researched article or paper.

Think of it this way: imagine you’re working with a researcher who hands off all their work without understanding the bigger picture of what you’re creating. That hand-off would never be seamless, and the result would feel disjointed.

This is exactly what’s happening with AI tools today—and it’s why we need to move toward agentic workflows.

Agentic Workflows Are A Cohesive System For AI Collaboration

So what are agentic workflows? They are systems in which specialized AI agents work together in a collaborative, orchestrated way to accomplish a single goal. Think of them like a well-coordinated team of specialists, each with a clear role and a shared mission. These AI agents aren’t just tools—they’re synthetic collaborators.

Here’s how it works:

• AI agents are assigned specific tasks based on their strengths. For example, one agent (like Perplexity) focuses on research, while another (like Claude) focuses on writing.

• An orchestration layer acts like a project manager, breaking down the user’s prompt into individual tasks and assigning them to the right agents.

• All context is logged and shared between agents, ensuring that nothing gets lost in translation. From initial research to final draft, the entire process is streamlined, with no need for messy hand-offs.

• The manager is you the human providing critical feedback and inspiration to redirect the AI agents work towards the output you envision.

It’s a bit like having a team of highly skilled, specialized professionals working together—except these professionals are AI agents.

Specialization Within Teams Leads To Highly Focused Output

The old adage "teamwork makes the dream work" couldn’t be more true because each team member brings their own strengths to the table. This diversity of perspectives leads to richer, more creative outcomes. So, why should AI agents collaboration practices be any different?

Specialization de-risks the creative process. Instead of relying on one generalist AI to handle every task, we can assign specific, highly capable AI agents to do what they do best. This specialization not only reduces the risk of mistakes (AI hallucinations), but it also allows for more focused, high-quality outputs.

By working together, these agents can generate better ideas and results than they could alone, just as teams of people do. The orchestration layer acts like a conductor, making sure every part of the system is working toward the same goal and that no agent is left in the dark.

Keeping Context Across Tasks Keeps Users In The Flow State

When working with agentic systems, context is king. The biggest challenge with fragmented AI workflows today is that context gets lost when you switch between tools. But in an agentic workflow, all actions, tasks, and user feedback are stored in a single, unified log. This means that the AI agents never lose sight of the big picture—and neither do you the human manager.

It also means you can stay in a flow state, that magical place where everything feels effortless and your productivity is at its peak. No more copy-pasting, no more reorienting yourself after switching tools. Everything flows seamlessly from one task to the next, with the AI agents constantly reacting as a unified team to you, their human manager. This leaves you to focus on the vision of the output as it is shaped from rough clay to a masterpiece.

Picture this hypothetical scenario for an agentic workflow:

To set the stage, you are behind schedule at work on your latest project and the assignment is due tomorrow! If this sprint isn’t completed on time it will delay the whole project’s completion. The team is blocked and to help them you need to produce a strategic research document.

• You are preparing to research a new topic at work. Instead of manually searching you decide to write a research thesis and task an AI agent team to conduct the research.

• You start by giving an AI a prompt, the research thesis, for a research-driven article.

• The orchestration layer creates a task cue and assigns tasks to the AI agent researcher for web research and AI agent copy writer for outlining and composing the article.

• Both agents know exactly what the end goal is, and they collaborate to build a polished final product.

• You provide feedback at key points in the process, helping the agents fine-tune their work, but never losing sight of the broader objective.

• The entire project—from the initial research to the final draft—is logged and documented in a single, transparent, familiar workflow.

This is the future of AI-driven work. A future where agents collaborate with other agents and humans alike, in the familiar processes we humans work.

The Future is Collaborative

As we move deeper into the world of AI, the systems that will stand out are the ones that embrace collaboration—both among AI agents and between humans and these agents. These Agentic workflows will empower human capabilities with the scale and precision of machines.

It is this collaboration that is the natural next step in the human-AI agent evolution, allowing us to harness the power of specialized agents working in sync, all while keeping us in the creative flow.

It’s time to leave behind the disjointed, copy-paste-heavy workflows of the past and embrace a future where AI truly works with us, not just for us. That’s the future we envision, now that’s left is to go build it!

Emerging Agentic News:

This week there have been some big announcements that serve as interesting views into the future of AI agents and a thoughtful lecture on agentic systems.

Agents AI

Released last week by Darmesh Shah, CTO of Hubspot. This is billed as a professional network for agents. His vision for the marketplace is. tocreate an app store of agents where they can be hired and collaborated with on projects.

Meta’s Project Orion

Zuckerburg announced project Orion in a keynote demo. It is their latest experiment from reality labs in the form of a small form-factor pair of augmented reality glasses that work with a small puck and a wrist sensor to control spatial navigation. it is not explicitly an AI product, but Zuckerburg did call AR the final form factor of AI. We feel the convergence of AI and AR is an exciting opportunity that feels very real with this announcement. I can imagine interacting with humans and virtually embodied AI agents to enhance collaboration opportunities.

Andrew Ng On AI agentic Workflows

This talk I have re-watched multiple times Andrew Ng was the lead of Google Brain, co-founder of Coursera, founder of DeepLearning.ai and has been a professor at Stanford University. His clarity of thought surrounding the architecture and technology advancements towards agentic systems is unmatched. This is a must watch lecture for anyone building or learning about agentic systems.

Llama 3.2 Models Announced Yesterday

The latest Llama models were announced, what makes them so impressive is they are open source with open weights able to be fine-tuned and customized for our products. The small model can be used for edge compute on-device scenarios. The Llama stack update is competitive to any closed, frontier model suite. This new suite of models is interesting news for teams building ai agents as it could serve as an an ideal foundation model needed to customize multiple agents for agentic systems.

What do you think? Are you excited about the potential of agentic workflows? How do you think this could change the way you work? Drop a comment—I’d love to hear your thoughts!