The Chat UI Trap: Why AI Products Are Regressing to 1980s

Featured Speakers:-

Links and Resources

• The Agent Architecture Decision: Workflow, Single Agent, or Multi-Agent
• From Demos to Deployment: Why MCP Is the Foundation of Agentic AI
• Marketing AI Agents: What They Are, How They Work, and Why Modern Teams Need Them

Transcript

[0:17] Host (Rajnish Ranjan)
Welcome to another episode of AI Machines and Marketer. Today’s session title is The Chat UI Trap — why AI products are regressing back to 1980s interfaces. Chat has quietly become the inseparable part of AI interfaces. It feels simple, familiar, and powerful. But the blank box also places a great deal of responsibility on the user: you need to know what you want, how to ask for it, and how to respond reliably. Is chat the best way to interact with AI — or are we just replacing buttons, dashboards, and menus with one modern version of a command line? To unpack this, please welcome Arpit Srivastava, VP of Growth Marketing and Analytics at Growth Natives, accompanied by Akhil Yadav, Associate Director of Product Management at Growth Natives.

[1:57] Arpit Srivastava
It feels like a co-host as guest — always excited to be here.

[2:04] Akhil Yadav
Really excited to be on my first podcast at Growth Natives.

[2:27] Host (Rajnish Ranjan)
Arpit, my first question: why has chat become the default interface in AI products?

[2:34] Arpit Srivastava
I think it’s like history repeating itself. If you go back to the 1980s and early 90s, we had command-line interfaces — operating systems where you had that blinking cursor. You had to have the command in your memory before you could do anything as simple as creating a file. So this is more like a DOS moment for AI interfaces. Whenever big shifts happen like this, engineering teams tend to be more heavily involved. From a product management perspective, it is a little debatable — probably it’s easy to get started with a chat interface versus thinking deeply about what the user expects. It’s a good start, but as applications evolve, the UI for AI will also evolve.

[3:54] Akhil Yadav
I also think that because this technology has been made easily accessible and there’s a lot of buzz around it, every company wants to be an AI-first company. So they are forcefully implementing AI chat — that’s also one of the reasons why it’s booming. There’s a tight deadline: you have to have an MVP in three weeks. That’s the mandate. Then obviously you’ll have to start somewhere, and having a chatbot gives the perception that you’re delivering. In fact, I’ve seen so many applications where it’s not needed — it actually adds friction — but they still have a chatbot.

[4:44] Host (Rajnish Ranjan)
Is chat a real step forward or a return to command-line thinking?

[4:53] Akhil Yadav
I think it’s both, depending on how you use it. On one end, it’s an incredible innovation — allowing users to interact with software in natural language for the first time. That’s a huge leap forward. But on the other hand, if you’re replacing a well-designed UI with a chatbot without thinking about the use case, it adds friction. The whole purpose of decades of UX innovation was to reduce the amount of things the user has to remember. If you remove that progress, all the responsibility comes back to the user. So it’s a double-edged sword: if used correctly, very helpful; if not, it adds friction. Clicks have now become 100 words.

[5:55] Arpit Srivastava
With all the advancement of models, you’ll also see — when working with Claude, for instance — that they give you cues and suggest the next best action. So in certain scenarios that is genuinely helpful. But you’ve got to have that purposeful design in mind, otherwise the blank box has no direction at all.

[6:45] Host (Rajnish Ranjan)
Where does chat make AI faster, easier, or more natural to use?

[6:54] Arpit Srivastava
In scenarios where the end user is exploring — they haven’t got a full picture of what they want to achieve, they’re in a research phase — the chat experience is a great place to go. But where you have a clear view of the end task you want to accomplish, a well-designed UI interface still holds an edge. Take a food delivery app like Zomato — you already know the kind of things you order, you have a clear navigation path. If you had to type everything about your preferences every time, that’s friction. You can have a chatbot for some kind of support, but forcing it in scenarios where a UI would be faster just creates problems. You’ve got to think from the end user’s perspective: is it needed? There could be situations for chatbot only, situations for interface only, and hybrid situations where both are available.

[8:45] Host (Rajnish Ranjan)
What do users lose when visual interfaces are replaced with a blank text box?

[8:54] Akhil Yadav
The biggest thing the user loses is discoverability. When a new user signs up, it’s very rare that they know exactly what the product is capable of. They generally skim through pages and menus and try to figure out what the product is about. But if you remove all that UI and replace it with a chatbot that just says “ask anything,” the user is confused. They don’t know what to ask. You’re left in the middle of nowhere. If I give a practical example: we were recently working on a marketplace for AI agents at Growth Natives. Initially we decided to have chatbots for all the agents. Then Arpit came to save us — it was his idea that instead of building chatbots for every agent, we should focus on the use case: is it adding value, or is it adding friction? Depending on that, we decided — UI-based, chat-based, or a mix. That’s how you lose discoverability if you don’t consider the use case.

[10:25] Arpit Srivastava
We actually discussed this in what we called war rooms. The approach we have to take is use case by use case — thinking from the user’s perspective with the product management hat first, then the engineering hat, and then arriving at the right UI. Still, human behaviour and user perspective matter above all.

[10:52] Host (Rajnish Ranjan)
Does chat reduce clicks while increasing the user’s cognitive load?

[11:01] Akhil Yadav
That’s one of the biggest misconceptions in product management — that reducing clicks improves UX. There’s a whole book about it called “Don’t Make Me Think” by Steve Krug. The goal should not be reducing clicks but reducing the effort to complete a task. Effort has two types: physical — the number of clicks or pages — and cognitive — the amount of thinking required, the number of things to remember. A chatbot reduces the number of clicks, but what about the cognitive effort? The user now has to think: what is this product capable of? How should I phrase my requirement? What context should I provide? Is the product even able to understand my request? All that thinking is now shifted to the user instead of the product. That’s why chatbots are reducing clicks but adding cognitive load — and the years of evolution we did to make software easier to understand should not be removed just because AI can chat.

[12:39] Arpit Srivastava
Amazon started with the whole concept of single-click purchase — they even copyrighted it, and Apple bought that copyright. It was such a significant innovation. What evolved out of that was conversion rate optimisation: reducing clicks, reducing steps. Then two fields emerged side by side: experience optimisation and conversion rate optimisation. You have to strike a balance — ensure the user’s experience isn’t compromised while driving conversions. These principles apply equally to AI products. The problem is we’re in a rush to deliver something AI-visible and we forget to apply those hard-won UX principles.

[14:25] Host (Rajnish Ranjan)
How can a user discover AI capabilities when they don’t know what to ask?

[14:32] Arpit Srivastava
As models become more sophisticated, the ability to parse intent from a basic prompt will improve dramatically. You won’t be at the mercy of 100 perfectly worded words — the model will figure out your objective and backtrack to determine the steps needed. That’s what happens when you work with Claude on a complex task: you give it a high-level goal and it plans the execution, evaluates errors, and fixes them. You start with a simple objective and the model decides the path. That’s where things will become more solid over time.

[16:01] Akhil Yadav
Even a few chatbots already give you suggestive prompts — after a result, here are the next available options. That’s one way to help users discover what the product is capable of without having to think of everything themselves. The product is doing the cognitive work on the user’s behalf, which is exactly what good UX should do.

[16:26] Host (Rajnish Ranjan)
Why does chat struggle when workflows are repetitive, standardised, and reliable?

[16:34] Akhil Yadav
Let me explain with an everyday example. I commute to work every day. Once I sit in my car, I don’t realise when I’m turning the steering wheel or applying the brakes — it’s in muscle memory. I arrive at the office without thinking about it. But if I replaced that steering wheel with a voice command system, where I need to issue a voice command for every turn and every control, my physical effort would decrease but my cognitive load would make it impossible to relax for a single minute. The same thing happens in software when chatbots are applied to repetitive tasks. For open-ended tasks — writing a blog, creating a prototype, brainstorming — chat is fine because you know the goal but not the path. But for repetitive tasks where you know exactly what needs to be done, chatting through the inputs adds friction and increases time on task. Take Jira: I have all my inputs from stakeholders, I know my developers’ availability, I know what’s required. It’s far faster to have the board in front of me and drag and drop than to type everything into a chat.

[19:25] Host (Rajnish Ranjan)
What does the future of AI interfaces look like?

[19:37] Arpit Srivastava
The way we are thinking about our products is to decouple the thinker from the doer. Think of a skilled chef — they take the order, they don’t do all the work themselves, they route it to the right person, and when everything is done, they taste it and ensure the quality meets the standard. The same will apply to AI interfaces. You’ll have utilities that are deterministic in nature, AI utilities that are probabilistic, and an orchestrator — the thinker or supervisor — who understands the whole ecosystem. Based on what the user is asking, it will pull in the right tool. From a UI perspective, the way it should happen is that prompting and doing are shown side by side: on one side you’re interacting, on the right side you see things happening in real time. We’ll have to come up with similar patterns for nuanced applications.

[21:39] Akhil Yadav
I’m very doubtful that the future of AI interfaces will be chat-first. I think we are going to chat much less than we expect — not because AI will be less capable, but because AI will be much more integrated with our existing UI interfaces. Nobody opens a chatbot to correct a spelling. Spell-check works right there, in context, invisibly. Similar things will happen across software. AI will understand context, bring insights without being asked, recommend actions, and automate work. Conversations will still be there for open-ended tasks, but chatbots will be less prominent. The future is not software that talks like a human. It’s software that understands context so well that we don’t have to keep talking.

Rapid Fire Round

[23:15] Host (Rajnish Ranjan)
Arpit — what is one AI feature you wish existed as a button?

[23:24] Arpit Srivastava
Can I have two? First: a button that transfers the context of a chat from one model to another. We hit this constantly — Claude’s token limit is reached, and I want to continue in OpenAI. That interoperability would be incredibly useful. For certain use cases a given LLM is also more relevant — Perplexity has strong context about company data, for instance. Being able to navigate between models mid-conversation would save a lot of time. Second: a clone button for an existing chat, so I don’t have to start again from scratch for a similar task. If you keep adding to the same chat it can get unwieldy — being able to clone it at a certain point and branch off would be very practical.

[24:38] Host (Rajnish Ranjan)
Would you rather explore a tool or watch a tutorial first?

[24:42] Arpit Srivastava
Neither — just explore. Jump straight in. That’s honestly how you learn fastest with any tool.

[24:46] Host (Rajnish Ranjan)
What is the most overrated part of prompting?

[24:50] Arpit Srivastava
Prompt engineering. When we started 2–3 years back, it was the hottest skill going. But I think now it’s becoming overrated — so much is happening at the model level that the need for highly precise prompt crafting is declining. It’s still relevant, but it’s on alert. The models are getting much better at interpreting intent from imperfect input.

[25:03] Host (Rajnish Ranjan)
Should an AI remember your working style automatically?

[25:08] Arpit Srivastava
Definitely — it should personalise. But it should not be creepy about it. Personalisation has been notoriously bad the last 10 years. The right approach: you should know why you’re seeing certain things, and you should have the ability to control and overwrite it if it’s not working for you. Transparency and user control matter just as much as the personalisation itself.

[25:27] Host (Rajnish Ranjan)
Which industry most urgently needs a better AI interface?

[25:32] Arpit Srivastava
There is no industry that remains untouched — there’s always an AI use case. But if I had to pick one: the supply chain industry. Most of my friends in supply chain are still using Excel sheets and complex formulas for critical daily operations. That’s one industry that really needs to embrace modern AI interfaces in a meaningful way.

[25:55] Host (Rajnish Ranjan)
Complete the sentence: The future of AI interface is…

[26:01] Arpit Srivastava
Intelligent enough to know whether the chatbot is required or not.

[26:10] Host (Rajnish Ranjan)
Akhil — what is the first thing you notice when you open a new AI tool?

[26:17] Akhil Yadav
Whether it helps me understand how to use the product — or whether it expects that I already know everything about it. That’s the deal-breaker point that determines whether I go forward or look for an alternative. Nobody wants to go through a full tutorial before using a product. That’s what AI was supposed to solve — making things simpler. If I’m doing manual onboarding work just to understand the tool, we’ve failed the purpose.

[26:40] Host (Rajnish Ranjan)
What is more frustrating: too many options or no visible options?

[26:44] Akhil Yadav
No visible options — always more frustrating. When you have too many options you can still scan and ignore the irrelevant ones. The human eye is actually very good at filtering. But when there are no options and you face a blank slate, you’re completely stuck. No direction, no starting point.

[27:00] Host (Rajnish Ranjan)
What is one task you would never want to complete through chat?

[27:0] Akhil Yadav
Managing my Jira board. Typing all my inputs — stakeholder requirements, developer availability, sprint priorities — into a chatbot would take forever. It’s so much easier to have the board in front of you, drag and drop tickets, and see the full picture at once. Same for creating a roadmap: you need to see everything laid out spatially, not typed out line by line. For visual, spatial tasks, chat is genuinely the wrong tool.

[27:19] Host (Rajnish Ranjan)
Would you trust an AI-generated workflow without reviewing it?

[27:24] Akhil Yadav
Not as of now. Maybe in future — but we’re still a long way from that. The moment you remove human review from AI-generated workflows, you’re introducing compounding risk. Human oversight is not optional yet.

[27:30] Host (Rajnish Ranjan)
What is more important in an AI product: speed or control?

[27:35] Akhil Yadav
Control. Most AI products today have already sped up the underlying processes — speed is table stakes now, it’s everywhere. But control is going to be the real differentiating factor. The products that give users genuine visibility into and control over what AI is doing and why will win. Speed without control just means faster mistakes.

[27:50] Host (Rajnish Ranjan)
Which traditional interface element should AI products bring back?

[27:58] Akhil Yadav
Dashboards. A well-designed dashboard gives you a rich view of your operations and surfaces insights without being asked for them. That’s exactly what AI should be doing proactively — and it’s something most current AI products are missing. The combination of AI intelligence and a well-structured dashboard is genuinely powerful and underexplored.

[28:22] Host (Rajnish Ranjan)
That’s it for our session today. Thank you so much, Arpit. Thank you, Akhil. It was really great talking to you both. Today’s discussion reminds us that the future of AI may not be about choosing between chat or traditional interface — it may be about designing the right combination of conversation, visual controls, automation, and human oversight. The best AI interface will not only respond to the user, but help them understand what is possible, remain in control, and complete the work reliably. Thank you for joining us. Until next time — keep learning, keep exploring.

See more webinar