I spend a lot of time experimenting with AI tools. Maybe not as much as Matt Berman, but a lot. And in the last six months, something has changed. AI isn't just helping me get more done. It's changing how we operate Forward Future, from the amount of work we can take on to the quality of what we produce.
We built a Codex workflow that handles much of the repetitive operational work behind our daily newsletter, from researching stories and organizing our selections to formatting the issue in Beehiiv. What used to require more than three hours of my time now takes about 20 minutes, freeing me up to focus on what matters most: editorial judgment, original analysis, and writing content worth reading.
I've also been building what I call my second brain, an AI assistant that connects my email, meetings, CRM and other business systems. It prepares me for meetings, keeps track of commitments and watches for things that need my attention. A few weeks ago, it caught an incorrect link in an email to a partner without me asking it to check. Small save, but one I might otherwise have missed.
Today, the greatest leverage comes from figuring out what I no longer need to do myself. The models are getting smarter, but connecting them to the right information and giving them useful responsibilities is where I've seen the biggest payoff.
Over the past three weeks, ChatGPT has introduced a bunch of features that make this all easier.
I picked ten worth knowing about. For each, I'll explain what's changed and how I'd put it to work, drawing on things we're already doing at Forward Future and a few experiments I'd like to try next.
Not everything is available to everyone yet, so I've flagged the important limitations.
1. Give ChatGPT a Screenshot and Let It Fix Things
New feature: GPT-6.1 Terra — October 26
I recently spent too much time working on diagrams for an article about my AI second brain.
The diagrams weren't particularly complicated. One showed how my assistant connects to its memory database and the applications it uses. Another illustrated how a company record links to contacts, agreements and outstanding commitments.
I wanted them to look like Forward Future, not the default diagrams ChatGPT tends to produce. So I kept going back and forth: simplify the design, change the colors, remove the background, preserve the content, fix the alignment.
All relatively small changes. But describing exactly what needs to move, by how much and what shouldn't change can become its own project.
That's what makes GPT-6.1 Terra interesting. The model can interpret what's happening on a screen and interact with software to complete tasks. Instead of endlessly describing a visual problem, you can show it what you're looking at and ask it to work through the fix.
I'd start with three applications:
Fix visual inconsistencies. Give Terra a screenshot of a page or graphic alongside a reference. Ask it to compare typography, spacing, alignment and hierarchy, then make corrections without changing the underlying content.
Review something before publishing. Have it inspect a newsletter, landing page or ad creative. Look for text that overflows, missing images, inconsistent layouts and buttons that don't work.
Investigate a dashboard. Rather than manually navigating through a dozen reports to explain an unexpected change, ask it to inspect the dashboard, investigate the relevant metrics and prepare a summary of what it finds.
Here's a prompt I'd use for a design review:
Compare this page against the reference design. Identify differences in typography, spacing, alignment and visual hierarchy. Correct them without changing the copy or overall layout. Then inspect the finished page and tell me what you changed and anything you couldn't fix.
One lesson from building agents: completing an action and verifying that it worked are two different things. For anything that changes a design, document or application, I want the assistant to check the result before calling the job finished.
Availability: ChatGPT Work and Codex on eligible paid plans. Terra isn't available in the regular Chat tab.
2. Build an Assistant That Remembers What You Promised
New feature: iCloud Mail and Calendar plugins — October 27
One of the more useful workflows I've been experimenting with is something I call the Commitment Breaker.
The idea came from a problem I'm guessing most people have. I make commitments constantly, often in the middle of another conversation. I'll promise to introduce someone to a founder, send a proposal, review a document or follow up next week. Most of the important ones end up in Asana or my calendar. But not everything makes it there.
My assistant helps track those promises across email and other connected sources. And I've been working on the reverse, too: commitments other people have made to me (arguably even more important).
That second part is especially useful when you're managing partnerships or sales. Someone might promise to send over a contract by Friday, but Friday comes and goes. Unless you're actively tracking it, that commitment can disappear into an email thread.
I currently do this through an assistant connected to Gmail and Google Calendar. ChatGPT's new iCloud integrations give people who rely on Apple Mail and Calendar a starting point for similar workflows.
Here's an easy first experiment:
Review my emails from the past two weeks. Identify commitments I've made to other people, including deadlines, deliverables, introductions and follow-ups. For each, tell me what I promised, who I promised it to, when it's due and whether there's evidence I've completed it. Include a link to the original email.
Once that's working, flip it around:
Now find commitments other people have made to me. Identify anything overdue or approaching its deadline. Don't assume a commitment is incomplete just because you can't find the follow-up. Flag anything you're uncertain about.
I'd run both manually before turning them into recurring routines. You're trying to teach the assistant what actually constitutes a commitment, which is harder than simply finding emails containing words like "Friday" or "I'll send."
You can also use the new integrations for meeting preparation. Ask ChatGPT to review tomorrow's calendar, find relevant email history and identify anything you need to prepare or resolve before each meeting.
Availability: iCloud Mail and Calendar plugins are available across ChatGPT plans on supported platforms, subject to account and workspace permissions.
3. Stop Building Presentations From Scratch
New feature: ChatGPT Slides — October 16
At Forward Future, we make presentations for all sorts of things: sponsors, potential partners, educational content and team meetings.
Creating slides isn't the hardest part. It's deciding what information deserves a slide.
Take a sponsorship proposal. We might have audience demographics, newsletter performance, YouTube engagement, previous campaign results and a dozen different sponsorship opportunities. All useful information, but throwing it into a presentation doesn't automatically make a persuasive pitch.
ChatGPT Slides can now create editable presentations, work from reference decks and support collaboration through shared Spaces.
Two things I'd recommend trying:
First, give it a presentation you already like. This is something I've learned from working with AI on other creative projects. Showing it the visual direction you want is often easier than describing every design decision.
Upload a reference deck and ask it to follow the typography, colors, spacing and layout conventions while replacing the content.
Second, give it the underlying material and let it help develop the story.
For a sponsorship proposal, I'd upload our audience data, previous campaign results and information about the prospective partner, then ask:
Build a sponsorship proposal using these materials. Before creating slides, identify the five most compelling reasons this partnership makes sense and propose a narrative for the presentation. Use the reference deck for visual direction, include only verified figures and flag anything important that's missing.
I'd review the proposed narrative before asking it to build anything.
There's little value in having AI produce a beautiful 20-slide presentation if you've organized the argument around the wrong information. Once the story works, the design becomes much easier to refine.
Availability: Eligible Pro, Business and Enterprise users with Space access.
4. Put Multiple Agents on the Same Research Problem
Keep Reading
Available to paid subscribers
Get the Frontier Pass to access Forward Future's private Slack community, member-only videos and content, insider interviews, and monthly AMAs with Matthew Berman.
Join The Frontier Today
