Agentic AI

General-purpose AI agents demonstrate promise (and disappointments)

With the recent release of Perplexity's Comet Browser and ChatGPT Agent, you may have been hearing more about AI agents.

"The big jump in AI over the past year has been moving from AI that you chat back and forth with to get work done, to a paradigm of AI Agents that you can send tasks off to and they go and complete work for you. Agents can go off for hours and complete tasks for us, and will just notify us when they're done doing their work."
- Aaron Levie, CEO of Box, on agents

For general-purpose AI agents like Perplexity Comet and ChatGPT Agent, it is still early days, and by all reports, these agents' performance is inconsistent at best:

"It’s 2025, and we’re sitting at the edge of what could become a watershed moment for artificial intelligence: the quest for a general-purpose AI agent. This isn’t merely a technical curiosity. It’s a vision that could reshape entire industries, transform daily workflows, and redefine productivity. But—and here’s where the stakes become crystal clear—vision is not reality. Yet."
- Nate Jones on his (paywalled) Substack

What type of tasks might general-purpose agents automate? Email triage and response, meeting prep and follow-up, and generating status reports by aggregating information from multiple tools. In the consumer domain, they purport to automate tasks such as travel planning, online shopping, and meal planning.

Note: General-purpose agents are less suitable for specialized tasks, such as design and coding workflows; these are best supported by vertical (or domain-specific) agents.

Currently, Perplexity Comet requires a $200 monthly Perplexity Max subscription, and ChatGPT Agent requires a minimum ChatGPT Plus subscription ($20/month, with some usage limits). A great task to try with ChatGPT Agent is creating a PowerPoint presentation from some source material. If you're unable to try one of these agents out for yourself, check out the walkthroughs below.

Testing Perplexity’s new AI browser Comet
☕ Medium Read (6 minutes)

Demo: Creating a PowerPoint with ChatGPT Agent
Watch time: 2 minutes

Agentic AI

What organizations often miss when building AI agents

Above, we've talked about general-purpose AI agents. In the meantime, companies are moving forward with specialized agents that automate tasks within their organization. Armand Ruiz maintains that most companies, in building AI agents, skip the most important step: designing the standard operating procedure (SOP).

The SOP stage includes:

• Translating messy business logic into structured steps

• Capturing the tacit knowledge hidden in teams

• Building something reliable, repeatable, and scalable

Ruiz further shares LangChain's six-stage framework for developing production-grade AI agents, which includes the SOP stage:

1. Identify a realistic agent use case.

2. Design the SOP.

3. Build prompts that handle the core reasoning.

4. Connect to live data sources and build the orchestration logic.

5. Evaluate the agent; iterate to improve reliability and performance.

6. Deploy and monitor; continuously refine based on usage patterns.

The LinkedIn thread contains over 90 comments worth reading through, as people share their real-world experiences building agents.

LinkedIn post
⚡ Quick Read (2 minutes)

Broader Impacts of AI

The end of content as we know it (and what comes next)

I've always encouraged UX folks to pursue an understanding of business mechanics and strategy. Ultimately, we are designing for both users' goals and business objectives. Understanding business becomes even more critical as design becomes increasingly automated with AI — and AI disrupts existing business models. To that end, I encourage you to take on this longer read from Ben Thompson, "Content and Community."

Ben maintains that AI-generated content is revealing a stark reality: AI isn't just disrupting content creation—it's deconstructing it entirely. We're entering a phase where web content is completely commoditized, where AI can generate any content on demand. This development will unwind the current, largely advertising-supported publishing that has characterized the web to date. Ben’s proposed solution: rather than competing with AI's content generation, publishers should focus on building communities around shared experiences—something AI, with its individualized responses, fundamentally cannot do.

Ben's article is a fascinating read and well worth the time to consider its ideas.

Content and community
🔍 Long Read (18 minutes)

AI and the Design Process

AI prototyping flips the design process on its head

In the most recent episode of Dive Club, Michael Riddering (Ridd) interviews Nicholas Klene and Holly Lee from the Figma Make team. It's a superb discussion on how vibe coding tools like Figma Make are fundamentally changing design ideation and workflows. Highlights include:

• How Figma Make fits into your workflow

• Rethinking what “high fidelity” means for a prototype

• How code is the glue and what it means for collaboration

• Some less obvious use cases for prototyping with AI

Whether you're not yet using AI for prototyping or are experienced with it, this episode will provide some great ideas and inspiration.

Why AI has changed prototyping forever
Watch Time: 50 minutes

Building with AI

A step-by-step guide to successful vibe coding projects

At this point, I consider the ability to prototype using AI (aka vibe coding) as an essential skill for designers and product managers. If the above podcast episode has inspired you to get started or improve your vibe coding skills, the Mastering Lovable guide is a concise, step-by-step set of instructions to help you every step along the way. Though geared for using Lovable, its principles apply to any vibe-coding endeavor. I especially like its emphasis on planning before you prompt and how it presents four clear phases for any vibe coding project:

1. Lay the foundation.

2. Think in systems.

3. Build with precision.

4. Iterate and ship.

Mastering Lovable
💡Bookmark for reference

That’s it for this week.

Thanks for reading, and see you next Wednesday with more curated AI/UX news and insights. 👋

All the best, Heidi