If AI can do more work, will we necessarily know what to do with the time that remains?

When I speak of “becoming a more complete human,” I do not mean setting a standard of completeness for anyone. It is a reminder that, alongside learning to use tools, we need to care for learning, relationships, embodied experiences, and the responsibilities we choose to accept.

These things matter without first proving that machines will never do them. Human worth does not have to await the outcome of a contest between human and machine abilities.

TL;DR: Pichai and Hassabis did not jointly confirm that AGI would certainly arrive before 2030. This revision corrects the dates, statements, and uncertainty of their forecasts. AI may change work, but that does not guarantee liberation for everyone, make professional skills unnecessary, or establish that machines will never address physical activity or creativity. I suggest continuing to learn, clarifying questions, and valuing relationships and personal participation, while treating employment, distribution, and responsibility as practical issues that still need attention.

1. Put 2030 back in the category of forecasts

The original article claimed that Hassabis and Pichai had almost simultaneously confirmed in June 2025 that AGI would arrive before 2030, even describing this as confirmation of internal progress. That was inaccurate.

In his interview transcript published on June 5, 2025, Pichai estimated that it might be slightly after 2030, while expecting major progress across many aspects by then.[1] “Technology will make substantial progress” cannot be rewritten as “AGI will certainly arrive on schedule.”

As a subsequent addition, in his July 23, 2025 interview, Hassabis estimated roughly a 50% chance within five years, around 2030, and discussed questions of definition and consistency across capabilities.[2] This material appeared after the article’s original June 9 publication date; it cannot be retroactively presented as something he had said in June.

Both were personal forecasts at the time, not product delivery dates or outcomes verified in 2026. I do not use them to announce when AGI must arrive, or let a job title remove uncertainty from a forecast.

2. Learn to collaborate—and to check

I still like the idea of people and tools changing how work is done together. But comparing AI to an intern should not turn it into something all-knowing. What it completed, what data it used, and where it is uncertain deserve checking.

For me, everyday collaboration with AI calls for asking which tasks a tool can perform and which judgments remain the responsibility of an authorized person. Saving time is a possible goal, not an automatic result of adoption.

My suggestion is to begin with a task whose delivery can be assessed, and specify the output, data scope, and review method. Retain the ability to correct errors and stop work, rather than accept an answer automatically because it is fluent.

This is my working approach, not a productivity experiment completed for this article. It does not assume that everyone has the same resources or freedom of choice.

3. Good questions do not require declaring that AI can never ask them

A question can move from “Organize this information for me” to “What am I actually trying to understand?” That process calls for clarifying purposes, listening to other people’s circumstances, and being willing to revise an initial assumption.

I think it is worth practicing. But I will no longer defend its importance by saying that machines can never ask new questions. This article has no evidence for drawing such a permanent boundary around all future systems.

Likewise, strong performance on some tasks alongside errors on others can remind us to inspect differences in capability. There is no need to attribute “jagged intelligence” to an unverified originator, or guarantee that every gap will eventually disappear.

The agency I want to retain is the ability to explain what I am trying to accomplish, which outcomes I will not accept, and who needs to participate. That is my view of tool use, not a final experimental conclusion about human and AI creativity.

4. Personal experience matters without placing movement beyond machine research

Pottery, woodworking, gardening, and cooking can be activities people choose to participate in themselves. Their meaning in life need not depend on being “impossible to encode forever.”

As material published after the original article, Google DeepMind’s June 30, 2025 Gemini Robotics On-Device model card describes inputs including text, images, and robot sensor values, and outputs including actions. It also lists risks involving hallucinations, environmental robustness, and inconsistency with instructions.[3] Embodied manipulation is already within AI research; it cannot simply be declared territory that machines cannot touch.

That does not establish that machines have complete human bodily experience, or can replace every craft. The model card is the developer’s account of a specific system; I conducted no independent operational test.

I would rather ask whether you want to spend time on something, experience the process, and complete it with other people. That is a life choice, not a neuroscientific effect or occupational guarantee proposed in this article.

5. Keep learning skills while making room for relationships and responsibility

“Becoming a more complete human” should not mean abandoning professional skills. Learning to interpret information, understand work contexts, recognize errors, and use AI can proceed together.

I would make a learning goal concrete: choose a problem that currently needs understanding, practice explaining it yourself, then check the tool’s answer. This is my suggestion, not the only list of skills that will never become outdated. I am not predicting that skills will necessarily expire in five years.

Relationships also deserve attention. Understanding someone, keeping a commitment, and working through disagreement need not first be priced as future scarce assets to be worth doing.

When work changes, income, training opportunities, labor arrangements, and responsibility do not resolve themselves. Individual curiosity matters, but structural issues cannot all be handed back to a person with the instruction to “try harder to be yourself.”

6. We can choose how to participate without waiting for AGI

The original article presented “machines take over work, and people receive their calling” as a natural outcome, and reassured readers not to worry about replacement. Those visions are not guarantees; I withdraw that certainty.

I still hope tools can reduce repetitive burdens and give people more time to learn, care for each other, and do things they find worthwhile. Getting there also requires work design, resource distribution, public discussion, and concrete arrangements for responsibility.

For individuals, small choices can begin today: learn something that requires understanding, clarify a question, fulfill a commitment with care, or participate in a process yourself. These are directions I value, not qualifications for survival after AGI arrives.

For me, becoming a more complete human means continuing to ask what I care about and how I am willing to participate, rather than proving that I cannot be replaced.

Four-panel comic: If AI can do more work, will we necessarily know what to do with the time that remains? learn something that requires understanding; clarify a question; fulfill a commitment with care; participate in a process yourself; These are directions I value, not qualifications for survival after AGI arrives.

Sources and scope of reading (checked through November 4, 2026)

  1. Sundar Pichai interview transcript, Lex Fridman #471: episode published June 5, 2025; approximately 00:39–00:42, especially 00:41:24. The host’s transcript warns of possible transcription errors. This was a personal forecast at the time, not a corporate delivery date.
  2. Demis Hassabis interview transcript, Lex Fridman #475: episode published July 23, 2025; approximately 00:52–00:57, especially 00:52:33. A later addition relative to the original article; probability, definition, and capability conditions are retained rather than treated as an observed outcome.
  3. Google DeepMind, Gemini Robotics On-Device model card: June 30, 2025 version; inputs and outputs, purposes, and safety limits across five pages. Published after the original article and used as subsequent material; a developer’s description is not independent verification or evidence of universal replacement capability.