Activity
Mon
Wed
Fri
Sun
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
What is this?
Less
More
3 contributions to The AI Advantage
From Building Context to Reclaiming Time: My First AI Capacity Milestone 🚀
🚀 Today, I celebrated a milestone: the context I’ve been building with AI started giving me time back. One principle stayed with me as I put what I learned in the IA Advantage bootcamp into practice: AI increases capacity. The gold metric is time reclaimed. For a while, I could feel the difference more clearly than I could measure it: less friction, less rebuilding context, more continuity. Tracking minutes saved didn’t fully capture what was changing. Then today, something clicked. A work block I had budgeted 75 minutes for was completed in approximately 4—including consulting sources, consolidating information, creating the deliverable, and closing the loop in Calendar. Those four minutes were built on a lot of prior work. I’ve been developing what I call “executable memory”: making context, decision criteria, and sources available so AI can use them to carry work through to completion. But there was also a golden rule: Freed capacity is never automatically reassigned. Any new allocation requires a fresh decision. Finishing early no longer meant immediately filling the gap with another task “because there’s time now.” A second rule helped protect that space: Spotting an opportunity does not authorize spending capacity on it. We could notice ideas and improvements without turning every discovery into more work. I believe these boundaries helped make today’s result possible: less scope expansion, less context switching, more accumulated context, and greater autonomy for AI. I wouldn’t use this single result to reduce future estimates yet. I want to see whether I can repeat it consistently. But today, I could finally put a number to something I had been experiencing qualitatively: the work of building memory was starting to return real capacity. So I put a sticky note on my desktop to celebrate. 😊 My takeaway: sometimes the impact becomes visible in how work feels and flows before it shows up clearly in the numbers. Staying with that principle mattered.
From Building Context to Reclaiming Time: My First AI Capacity Milestone 🚀
0 likes • 19d
@Tam Hn Exactly, Tam! 😊 And an added benefit I keep noticing is how much this carries over into working with people. Defining what AI can decide, where my judgment is needed, and when to stop makes me examine how clearly I communicate those boundaries in the first place. Learning to set them thoughtfully is meaningful work in human relationships, too!🙃
1 like • 19d
@Rene B Thanks, Rene! 😊 I’d be happy to share what I find when I review that first completed run. Your questions have helped me clarify both how the system fits together and what to look for in the records. I really appreciate the exchange!
⚖️ The Sweet Spot Is Seven to Ten Percent of the Week
Many of us think the way to win time back with AI is to use it for as much of the working day as we can. More tools open, more tasks handed over, more hours spent inside the assistant. But the more meaningful shift is that the people who report the strongest productivity are not the heaviest users. They sit in a narrow band. In a large workplace study from ActivTrak covering more than 160,000 employees across over 1,100 organisations, people who spent roughly seven to ten percent of their working hours inside AI tools showed the highest productivity. Yet only about three percent of AI users fell inside that range. On a forty-hour week, the sweet spot is somewhere between two hours and forty minutes and four hours. That matters because it changes the time question. The goal is not to maximise the hours AI touches. It is to make a small, deliberate slice of the week do the most work, and to stop the rest of the week from quietly filling with prompts, retries and checking. ------------- Context ------------- Most of us have absorbed a simple story about AI and time: the more we use it, the more time we save. It is an easy story to believe because the early wins are real. A first draft appears in two minutes. A summary arrives before the coffee cools. So we reach for the tool again, and again, and the habit spreads across the day. The hidden problem is that use and benefit do not rise together forever. Every extra task handed to an assistant brings a small bill with it: writing the instruction, waiting, reading the answer, deciding whether to trust it, fixing what is off. For a task that would have taken ten minutes by hand, a twelve-minute round trip through the tool is a net loss, and nobody records it because the tool felt busy and helpful the whole time. This is where a different lens helps. Instead of asking how much we use AI, we can ask what share of the week is doing the real lifting. The ActivTrak finding suggests the answer is small. Seven to ten percent of the week, spent on the right tasks, beats a third of the week spent everywhere.
⚖️ The Sweet Spot Is Seven to Ten Percent of the Week
2 likes • 21d
This landed at exactly the right moment for me. Today I had a small but very tangible experiment with the other side of this equation. A work block I had budgeted 75 minutes for was completed in about four, because the context, sources, decision criteria and stopping point were already in place. But the part I’m finding most important is what happens after that time is reclaimed. I’ve started using a rule: freed capacity is never automatically reassigned. It goes back into a pool, and any new use of it requires a fresh decision. Otherwise, AI simply allows more work to rush in and occupy the space it just created. Your point about not every task deserving an “AI pass” also resonates with something I’ve been noticing. Sometimes using AI for a two-minute task actually adds coordination cost — and can even stop us from exercising skills we still need. At the same time, there are tasks where the comparison isn’t “AI versus two minutes of manual work,” but “AI versus the real cost of reaching the required quality.” For me, writing this comment naturally in English is a good example: the language, tone and idiomatic differences would make doing it alone substantially more expensive. So I’m starting to think about AI use through two questions: Does it genuinely reduce the cost of reaching the quality I need? And once capacity is reclaimed, do I actually protect it? Your post gave me a much sharper frame for both. Thank you.
🤝 Human-in-the-Loop Is Not a Safety Feature, It’s a Skill
“Put a human in the loop” has become the default answer to AI risk. It sounds reassuring, responsible, and complete. But in practice, simply inserting a human does not guarantee better outcomes. Without the right skills and conditions, it often creates a false sense of safety. ------------- Context ------------- As AI systems become more capable, many organizations rely on human-in-the-loop approaches to maintain control. The idea is simple. AI produces an output. A human reviews it. Risk is reduced. What actually happens is more complex. Reviewers are often overwhelmed by volume, unclear about what to check, and uncertain about how much responsibility they truly hold. Over time, review becomes routine. Routine becomes trust. Trust becomes complacency. This is not a failure of people. It is a failure of design. Oversight is treated as a checkbox instead of a practiced capability. Human-in-the-loop only works when humans are equipped to be there meaningfully. ------------- The Illusion of Oversight ------------- Many review processes look solid on paper. A human approves. A box is checked. A log is created. From the outside, risk appears managed. Inside the process, the reality is different. Reviewers face time pressure. Outputs often look plausible. Context is incomplete. The easiest path is to approve unless something is obviously wrong. AI systems are particularly good at producing reasonable-looking answers. That makes superficial review ineffective. When errors are subtle, humans miss them, especially at scale. The illusion of oversight is dangerous because it delays learning. When mistakes eventually surface, they feel surprising and systemic, even though the signals were there all along. ------------- Judgment Fatigue Is Real ------------- Human-in-the-loop assumes humans can sustain attention and discernment indefinitely. That assumption breaks quickly. Reviewing AI outputs is cognitively demanding. It requires holding context, spotting inconsistencies, and questioning confident language. When volume increases, fatigue sets in. Review quality drops.
🤝 Human-in-the-Loop Is Not a Safety Feature, It’s a Skill
1 like • Feb 3
Human-in-the-loop often exists in name when “review” is treated like a checkbox—but also when we ignore the emotional layer of supervision. In real life (neurodivergent/homeschooling motherhood, using AI mostly to think), cognitive fatigue changes everything: relief and trust can quietly replace discernment. That’s when HITL turns into rubber-stamping. What made the biggest difference for me is combining attitude + role clarity with three concrete indicators that define authority-based leadership: direction, order, and safety. They’re basically the three instruments on the dashboard that help you navigate an AI conversation: Direction: Who is setting the goal and constraints right now? Order: What is the structure of the interaction (steps, boundaries, decision points)? Safety: What risks are present, what needs verification, and when do we stop/escalate? Training supervision skills is crucial, and designing the control matrix matters too. But without explicitly naming these indicators, people don’t know what “good oversight” actually looks like moment to moment. So for me, HITL is “not in practice” whenever the human is overloaded, lacks real authority to stop/escalate, and is expected to supervise without signals, time, or training—especially under fatigue. Thank you for the article.
1-3 of 3
Marian Teves
3
28 points to level up
@marian-teves-8058
CEO of Peasymind Inc. & Executive Coach, helping parents and leaders to create more human, simple, and powerful learning experiences.

Active 9d ago
Joined Oct 9, 2025
Powered by