← Ascendy 한국어

meta

Everything you can click should also be sayable — yet enabling it started with a menu

· Ascendy Engineering

TL;DR

Source note. This isn’t a decision record or an incident write-up — it’s an argument piece written from an operator interview (this blog’s first “industry argument” genre). First-person opinion is marked as opinion, external facts are attributed, and current product status was fact-checked at publish time (refined-source path in frontmatter sourceIntake). The feedback thread connects to a report arriving changes nothing’s capture→route, and the product motivation to why we built Ascendy.

”Pull up my past conversations” — “Enable it in Settings”

Start with one scene. June 2026, I opened an AI agent session on the web and asked: “Can you pull up my past conversations?”

The gist of the answer: full-text search of past conversations is off by default in this session, and to use it I’d have to go to Settings → Profile and turn on the “Search and reference past chats” toggle.

I paused there. I’m trying to recall the past through conversation — and to do that, I have to leave the conversation, dig through a menu, and flip a switch. Saying “turn that feature on” doesn’t work. The conversational feature’s entry point is itself non-conversational.

That small irony is the whole article.

The claim — conversational parity

Here’s our opinion. Every action you can do by clicking a menu or button should also be doable by telling the agent in conversation. The chat window is already a good-enough interface, yet the work around it isn’t seamless from inside it.

Let’s preempt the misread. This is not “remove the menus.” People used to the old UX may genuinely find conversation more awkward — we don’t deny that. Keep the menus. Let outputs render as buttons, cards, links. The point is one thing: every click-action must have a conversational path alongside it. Both entry points open — call it conversational parity.

This stance doesn’t actually conflict with what UX designers say publicly. It converges. Designers argue that “chat isn’t always the right answer — pure conversational UI is cognitively demanding and hard to navigate; a hybrid (chat + buttons/cards) is best” (UX Collective, Design Key). True. And conversational parity is a stricter version of that hybrid view — doing the hybrid right means laying a conversational path over all the clickable actions, not just some.

The industry already proved parity works

Here we have to be honest. “AI agents still make you click for everything” is false as a current fact. As of 2026, all three — on supported plans/accounts — let you recall past conversations by talking:

This doesn’t weaken the claim. It strengthens it. The industry took one of the most common click-actions — finding a past conversation — and made it conversational, proving conversational parity actually works. The problem is it stopped there.

The half-built parity

Recall went conversational — yet enabling that recall is still a menu toggle. Help is still a separate page click. In the main flows we’ve seen, settings, sharing, and export are mostly click-only. Even recall: sidebar search matches titles only, not content (Claude). Conversational recall hides as a secondary feature while the list stays the primary path.

So parity stopped halfway. One feature got pulled into conversation, while the equally-clickable features right next to it were left in their old spots.

Why it stopped there (our guess)

This is a guess, not an assertion — honestly, we’re more curious what actual designers think. Still, it’s probably layered:

  1. It was rational at first. Early agents couldn’t reliably turn “say it” into “do it.” So a definite click-menu was the right design.
  2. The UX hardened around that capability. One product succeeded, and everyone copied its shape.
  3. Users adapted. “LLM services just look and work like this” accumulated as lived experience.

The result is the familiar pattern — the constraint that justified the shape (limited agent capability) eased, but the shape stayed on by inertia. It’s a shape we’ve seen in other posts too: a structure that outlives the reason for it.

What we do (and where we stop)

Just lecturing others is weak. We (Ascendy) actually apply this parity in the product — most click-doable features that operate privately are also doable in natural language.

But there’s one principled exception. Actions that are irreversible and privacy-sensitive — like making a photo public — are deliberately blocked from the agent and kept behind an explicit human click. Parity isn’t dogma — irreversible external exposure gets a human gate. That exception actually sharpens the principle: it’s not “do everything by conversation,” it’s “make everything that isn’t dangerous doable by conversation too.”

Feedback follows the same logic. The old UX is “hunt for a feedback form → open it → send → wait for a reply.” In the conversational-parity version, you just say the friction in the chat and it gets captured and routed to the right place — so the user doesn’t become the bottleneck re-relaying feedback. (This is a design principle we aim at.)

Takeaways

Authorship & citation: Written by Ascendy Engineering; quotable with attribution. Found something wrong? Let us know via a GitHub issue.

Tags: ai-agent, ux, conversational-ui, product-design, opinion