The future of work is voice, with apps that keep you in control

Key takeaways

+

• When AI agents do the busywork, a person's job shifts to directing and deciding, and talking is the fastest way to direct.
• Speech input on a phone is three times faster than typing, and a spoken request is often one sentence where the work itself is a dozen clicks.
• Chat is a good place to ask and a poor place to review. Work has state, and checking an agent's work needs a screen built for it.
• Agentic apps serve two users at once: the agent does the work through the app's tools, and the person sees what changed, what's waiting, and why.
• Most clicks inside a company's apps will soon come from agents, while people stay in control through permissions, approvals, and a record of every action.

For forty years, working with software has looked about the same: sit down, open the right app, and type. The keyboard and the mouse were built for people who do every step of the work themselves.

AI agents change who does the steps. When an agent can research a company, update the CRM, draft the follow-up, and book the meeting, your job moves up a level: decide what should happen, then check that it happened right. For that job, your fastest interface is your voice, and your most important one is still a screen.

Here's the bet we're making at Opser. You'll direct your agents by talking to them. They'll do most of the clicking inside your apps. And your apps will show you what got done and what's waiting for your call. You stay in control, and the busywork goes to the agent.

Why is voice becoming the interface for work?#

Talking is the fastest way most people have to say what they want. In a Stanford study, speech input on a phone was three times faster than typing, with fewer errors. And intent is short when you say it out loud. "Move the Northstar call to Thursday and tell Robin" is one sentence. Doing it yourself is a calendar, an email, a CRM update, and a dozen clicks.

Voice assistants have been around for years and never made it into real work. They could set a timer or read the weather, and that was the limit, because they had no access to your customers, your calendar, or your team's work, and no way to act on any of it. Agents close that gap. An agent works inside your apps with your permissions and your company's context, so a spoken request can finally become finished work.

Voice also reaches the parts of the day a keyboard can't: the walk between meetings, the drive to a customer, the minute after a call when you remember the three things you promised. Those are the moments work usually waits. With an agent listening, they become the moments work gets assigned.

Why was chat never the right interface for work?#

Chat made AI usable for everyone, and it's still the right place to ask a question or hand off a job. It's a poor place to check the result.

Work has state. A pipeline has stages, a calendar has conflicts, an inbox has threads waiting for an answer. In a chat, every result arrives as one more message, and the thing you approved twenty minutes ago has already scrolled out of view. Reviewing thirty outreach emails one message at a time, or piecing together which records an agent changed this morning, takes longer than doing the work yourself.

The more an agent does, the more you need a place to see it: what changed, why it changed, and what's waiting on you. That place is an app.

What is an agentic app?#

An agentic app is built for two kinds of users at once. Your agent uses it through tools, to search, create, update, and draft. You use it through an interface designed for directing and deciding: the queue of work waiting for your OK, the evidence behind each item, and the record of what's already done.

You can see the pattern across Opser's apps:

• In Sales, every message the agent writes shows the evidence it used, and by default nothing goes out until you approve it in the Master Inbox.
• In Content, drafts land in Ready with proposed posting times, and any fact the agent couldn't confirm comes back to you as a question.
• In Meetings, a call becomes notes, decisions, and action items, and the agent drafts the recap for you to send.
• In every app, a sensitive action stops on an approval card, where you can allow it, deny it, ask for an edit, or always allow a routine one.

Each of these screens answers the same three questions: what did the agent do, what's waiting for me, and what is it based on?

Will AI agents use your apps more than you do?#

Count the clicks in a normal week: moving deal stages, logging calls, updating tasks, chasing invoices, filing documents. Most of it is mechanical, and it's exactly what agents are good at. As they take it over, most of the activity inside a company's apps will come from agents. People will open those apps less often, for shorter and more important moments: to decide, to check, and to correct.

That raises the bar for software. Every action an app offers has to work for an agent, under the same permissions a person has. Every change needs a record of who made it and for whom. And the screens people still look at should lead with decisions: what needs approval, what changed since yesterday, what's stuck. In Opser, every app gives your agents its tools, and every change and send lands in an activity log you can review.

How do you stay in control when agents do the work?#

Directing by voice only works if you trust what happens after you stop talking. Three things build that trust:

• Permissions. Your agent works with your access, and a shared agent with the access you gave it, so it can't reach anything you couldn't.
• Approvals. Research and drafts happen on their own. Sends, deletes, and admin changes wait for your OK, on screen or out loud.
• A record. Every change and send is logged with who it was for, so you can always see what happened.

On a voice call, the agent describes the action and asks you to say approve or reject, and an unclear answer runs nothing. Approval is the moment you stay in charge, and a good agentic app makes that moment take seconds. Read more about how approvals and permissions work in Opser.

How can you work by voice today?#

You can start now in Opser. At your computer, open the chat with your agent and start a live voice call, or press Cmd+Shift+Space. Talk the way you'd brief a colleague. Your agent answers out loud and works in your apps while you talk, and the call stays with you as you move between apps, so you can watch the deal update while you're still talking about it.

On your phone, open your workspace in the browser or in the Opser app for iPhone and iPad, and the same agent is one tap away for the walk between meetings. More on that in Mobile & voice.

A morning might sound like this:

What's on my calendar today, and which deals haven't moved in two weeks?

Draft a follow-up for each of the quiet ones, and move my three o'clock with Daybreak to Thursday.

Read me the Daybreak email before you send it.

The agent drafts, asks before it sends or moves anything, and you say approve. Later, at your desk, your apps show you everything it did.

The interface for work is splitting in two. You'll say what you want in a sentence, wherever you are. Your agents will do the clicking. And your apps will become the place where you see the work and make the calls. That's the workspace we're building at Opser: agents for the busywork, apps that keep you in control, and your voice to direct both.

Frequently asked questions

Will voice replace the keyboard and screen at work?

Voice will become the fastest way to direct work, and screens will stay where you review and decide. Most people will talk to start things and look to approve them.

What is an agentic app?

An app built for an AI agent and a person at the same time. The agent does the work through the app's tools, and the person sees what changed, what's waiting for approval, and what each item is based on.

Can you approve an AI agent's actions by voice?

In Opser, yes. During a live voice call the agent describes the action and asks you to say approve or reject. An unclear answer runs nothing, and the approval stays recorded in the chat.

Does Opser keep recordings of voice calls?

Opser keeps the text transcript of the conversation in the chat and doesn't retain the audio.

How much does voice cost in Opser?

Live voice uses workspace credits, billed per second, and you see the rate before your first call. The agent's own work is charged at its usual rates.

Simone Zamparini

Founder · Senior Software Engineer

Simone Zamparini is a founder and senior software engineer.