The paragraph that beat the dashboard.

I used to spend the first hour every workday in five different applications:

By the time I finished, I knew what was on my plate. I had also lost an hour I should have spent doing the work.

So I built a morning briefing. Now I walk into the office, send one message ("briefing") and ninety seconds later I get one paragraph. The paragraph tells me what I owe people, what I am waiting on, what landed yesterday, what's flagged, and what shifted overnight. The five tabs are still there if I need them. I almost never need them.

The frame is this. You used to walk into the office and check five tabs before coffee. Now you walk in and read one paragraph.

Here is what the system does. It pulls everything that happened since the last briefing (Monday's covers the weekend, Tuesday's covers Monday) from each of the source apps. Gmail. All five Slacks. Calendar. Granola. Notion. Then it groups what it found into sections an operator actually thinks in: things I owe people right now, things I am waiting on, things in motion, things that landed yesterday, flags I should know about, and a small set of things worth saving for later. The paragraph is laid out in those sections. I read it top to bottom in three minutes.

A few decisions I had to make that an operator should know about, because they shape what the system delivers.

First, triage before composition. The earlier version tried to write the briefing in one pass. It pulled everything, summarized, shipped it. That produced briefings that looked clean but described yesterday's task state. The fix was to triage first: process the new action items, the new commitments, the new flags, write them to the task list, then compose the briefing from the updated state. Costs me about thirty seconds of clicking-to-approve each morning. The briefing reads as of today, not as of yesterday.

Second, sliding window instead of fixed twenty-four hours. The first version pulled "last 24 hours" every time. That meant if I skipped a day (a conference, a weekend, a sick day), the previous day's signal was silently lost. The fix was to track when the last briefing ran and pull everything since then. Catches up across weekends and bad days without missing anything.

Third, group by status, not by source. The earlier version listed every input in source order: Gmail items, then Slack items, then Calendar items. That made me read every section regardless of what was urgent. The fix was status-grouped sections: things I owe right now go in one section, things waiting go in another, things landed go in a third. I can jump to "what's blocked" or "what's owed" without reading top to bottom. This is the single change that turned the briefing from useful into something that actually replaces the five tabs.

Two things that almost broke it. The email connector treated "last 24 hours" as a calendar-day filter, not a rolling window, so the briefing pulled the whole previous day instead of the actual window I wanted. A small filter step on the timestamps fixed it.

The other one was the voice rules. The ones that keep this from sounding like a consultant deck almost got skipped on one morning's run because the script classified the work as data aggregation instead of writing. The briefing that morning was unreadable. Now the voice rules run as a hard preflight on every single briefing.

If you want one of these for your business, here is what to ask the person who would build it.

Four things to specify. The source list (your inboxes, your calendar, your task system, your meeting notes, whatever you actually check in the morning). The scoring rubric (what makes something an action item versus an FYI; the operator decides this, not the AI). The output destination (a Slack message, an email, a Notion page, wherever you will actually read it). And the cadence (daily on demand, or a scheduled run that lands before you sit down).

A competent builder can stand this up for a real business in a few days, not weeks. The system that returns the morning paragraph matters more than the model under it. The work is in the structure, not in the AI.

Three things worth your time this week.

Google Shopping introduces Universal Cart

AI is becoming a place people buy, not just a place they ask. Google's Universal Cart now follows a shopper across Search, Gemini, YouTube, and Gmail. The same week, Shopify switched on AI storefronts for every merchant, so your products surface inside ChatGPT, Gemini, and Copilot at no extra cost. If you sell anything online, the question is no longer whether AI sends you customers. It is whether your store is readable to it.

AI Use at U.S. Businesses

The Census Bureau put a number on the gap. Fewer than one in five firms with four or fewer employees use AI at all, and among the businesses that do, more than half apply it to three or fewer tasks. Read past the headline noise: most of your competitors have not started, and the ones who have are barely scratching it. That is not a reason to relax. It is the size of the head start available to whoever moves first.

Nearly half of enterprise AI initiatives are on track to fail

More than four in ten of big-company AI projects are headed for failure, per a new survey of 467 executives. The cause is not weak models. It is the missing layer around them: who owns the output, what gets checked, what happens when it is wrong. Same call I made in the Build. The AI drafts, a human approves before anything ships. Start with one workflow that has a clear owner, not ten that have none.

When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning — constraints, edge cases, examples, tone — and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.

89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Try Wispr Flow free — works on Mac, Windows, and iPhone.

Start flowing free

The pitch word of the season, decoded, with the question to ask back.

You are going to hear "agentic commerce" and "agentic storefronts" a lot this quarter. Strip the prefix. "Agentic" just means the software takes an action on its own instead of waiting for a click. An "agentic storefront" means an AI can read your catalog and check a shopper out without that shopper ever visiting your site.

That is a real shift, worth understanding. It is also a sales word built to make you feel late. So ask the question that cuts through it: what does this do that my current setup does not, and who is accountable when it does the wrong thing on my behalf?

The answer sorts the vendors building something useful from the ones renaming what you already have.

See you Wednesday morning.

Reply with what you're working on. It lands in my inbox at [email protected]. I read every reply, and it tells me what to write about next.

Talk Wednesday.

Marc