I killed cold email. My AI told me to send.

Hey — and welcome to Sunday Signal.

On Friday I killed my cold email system. Nineteen straight weekdays of drafts sitting in a mailbox nobody opened. It was over.

On Tuesday morning my Chief of Staff, an AI assistant I built myself, opened its Morning Brief with this line:

"First Move: send the 8 cold email drafts in Gmail now."

The channel was four days dead. My own assistant did not know.

That is the boring part of the story. The interesting part is that the assistant was not the only one still working off the old plan. My calendar had nothing on it for the new channel. The reminder that did fire told me to call bars at 10am, an hour the same system had already flagged as the wrong time to reach bars. And the whole thing was designed to leave no record of what I actually did.

Three separate machines, all quietly executing yesterday's plan.

1. Another top-tier model went open weights this week
On August 17 Z.ai released GLM-5.2 Turbo under the MIT licence, joining the run of high-capability models you can now run yourself for the cost of the electricity.

Why it matters for you: Every few weeks another model that used to be a subscription becomes a file you can download. The lever is not which model you use. It is what stays useful across all of them.

2. DeepSeek shipped its flagship with a "how hard should I think" dial
V4 Pro left preview on August 12 and now takes an effort setting: low, high, or max. Cheap and quick for easy work, slow and thorough for the problems that used to be why you paid for the top model at all.

Why it matters for you: You have been given a control you did not choose. If you leave it on the default you are trusting the tool to decide, and the tool defaults conservatively. The people getting more out of the same product this month are the ones who actually turn the dial.

3. OpenAI announced it will keep none of your data on the frontier tier
On August 20 OpenAI committed to Zero Data Retention for its frontier API customers, plus a new safety layer that watches for patterns without anyone at the company seeing what you sent. Rolling out through September.

Why it matters for you: Retention was quietly the reason a lot of teams could not use these tools on real work. If your business kept a "no external AI" rule because of data risk, that rule is worth re-reading against what the vendors now actually promise.

The dropped resolution is almost never a willpower failure. It is a coordination failure.

You changed your mind, and one system in the stack did not get the memo, and it kept scheduling the old life until the new one wilted. Nothing on your calendar disagrees when you have a new priority. Your notes dashboards do not update because you had a hard conversation with yourself on Friday. Your AI assistant does not know until you tell it. Your reminders keep ringing for the dead thing until you cancel each one.

Retiring something is half the work. The systems that kept it alive will keep pushing it long after you decide it is over.

Here is exactly what happened on Tuesday.

I had booked myself a 10am calling block, my first replacement for the cold email I had killed. Forty-five minutes, five bars and cafes on the list. Simple.

Fault one: my Chief of Staff, the AI that reads my calendar and my mail and tells me what to do first each morning, still thought cold email was the plan. Its first sentence to me on Tuesday was to send eight drafts of a channel I had retired four days earlier. It was not lying. It just had not been updated.

Fault two: the timing was wrong, and it was wrong by data the same system already had. Every lead on the list carried a note saying the best time to call this kind of business is 2 to 4pm, between rushes. Nobody would call a bar at 10am. My reminder did.

Fault three: nowhere was there a place to record what happened. So when I asked myself on Wednesday whether Tuesday's block happened, I had no evidence either way. That felt like willpower. It was actually a missing form.

I did not make the calls. Strike one. The rule I built to hold myself to this — two missed Tuesday blocks and outreach dies entirely — is now sitting at one.

Honest half: the new block moves to 2pm on Tuesday and I have not yet done it. I am telling you this before I have proof any of it worked.

Week 13 of building in the open. What it actually looked like:

Audit the four places your old goals still live. Ten minutes.

Pick one thing you have decided to stop doing, or one thing you have decided to start. Something real, in the last month. Then walk through these four surfaces and ask what each of them still thinks.

1. Your calendar. Look at next week. Are the blocks and holds still describing the old life? Is the new one on there at all?

2. Your AI assistant, or your daily brief, or your dashboard. Whatever tells you what to do each morning. Is it still prompting the retired thing? If it is, the system prompt or the source data is stale, and the fix is usually one paragraph.

3. Your notes and lists. The projects folder, the running todo file, the pinned document you open every day. Which ones name the old priority as if it is still the priority?

4. Your reminders and alarms. Especially the recurring ones. Every ping that fires for a dead goal is a small vote against the new one. Kill them today, not next week, because they will keep firing while you decide.

If you find nothing to change on any of the four, either you have never retired anything or the audit is worth doing more slowly. In my experience it is the second one.

Reply and tell me which of the four surprised you. I am collecting the funny ones.

Fifteen Saturdays in. The arc so far: ship, constrain, build, compound, patch, subtract, ask, remember, price, staff, equip, doubt, show, keep, and now retire. Deciding to stop is the loud part. The quiet part is going around and telling every system you built about the decision.

The thing this taught me is smaller than it looks. Most of what I thought was follow-through was actually infrastructure agreeing with me. When it stopped agreeing, I stopped following through, and I thought that was character. It was scheduling.

If any of this was useful, forward it to one person who is still doing something they decided to stop.

Quietly, deliberately. One Saturday at a time.

— Kalpesh
Founder, Sun Ventures

P.S. Work with me. Everything above came out of building real things for real use.