Steve Jobs once called the personal computer a bicycle for the mind.

A tool that amplified human capability far beyond what our bodies or brains could do alone.

He was right.

But the metaphor had one constraint: you still had to pedal.

Most software has always worked this way. You open the app. You ask. You click. You search. You decide. The tool waits for you.

That was also the first era of AI.

The Last Year

ChatGPT made answers easy. It could summarize, brainstorm, explain, rewrite, analyze. It felt magical.

Then the cursor blinked.

You still had to know what to ask. You still had to move the work forward.

That changed over the last year.

The interface shifted from answer box to work loop: Codex, Claude Code, MCP servers, custom skills, scheduled tasks, agents running in parallel. 

Now they don’t just write better text, they can handle goals, inspect context, take steps, and come back with work.

The bicycle started pedaling.

What This Actually Looks Like

I feel this most as a solo founder.

I do not have much of a team. I do have agents.

Most mornings, one agent pulls together what changed overnight: email, Slack, calendar, meeting notes, loose threads. It does not give me a dump. It gives me the top three things that matter.

Another watches production logs, checks whether an error is a real bug, traces the likely cause, and starts a PR for me to review.

A third handles research before calls: company context, market notes, recent signals, likely pain points, what I should ask.

When a workflow repeats often enough, I turn it into a skill file. Then the next time, I do not have to remember the process. I invoke it.

That is the operating model shift.

Most Teams Haven’t Noticed

Most teams in 2026 are still typing into a box.

They've got a shiny new token budget, a new friend named Claude…but the work doesn't look materially different than before. Ask better questions, get better drafts, then back to the same inbox, same meetings, same forgotten follow-ups. 

Prompts are table stakes. Managed loops are hard, high-value, and where the next decade of leverage lives. Which loops can you trust? Which don't need babysitting? Which can actually replace a workflow you do every day?

Autonomy also changes the cost model. A chat answer is one event. An agent loop is a recurring system: gather context, reason, act, test, retry, notify. The unit economics matter.

The demo is still the trap.

A great answer in ten seconds demos beautifully. A long-running agent that quietly watches a customer thread, notices a stalled commitment, checks the right context, and taps you on the shoulder three days later?

Less theatrical. Much more useful.

Not Just a Worker. A Watcher.

• Tell me when this customer goes quiet.

• Tell me when a production issue looks real.

• Tell me when a project has a commitment with no owner.

• Tell me when nothing happened and that is the problem.

That last one is the part people miss.

A useful agent is not only a worker. It is also a watcher. It notices gaps.

It turns company context into signal without waiting for the perfect prompt.

The Keen Observer

Apple products always understood this. Speed was never the point - reduced mental load was. They anticipated, and they made the next step obvious.

That's the thesis I left to build on. Communication context the agent can actually act on, not just retrieve.

Agentic AI is bringing that product philosophy to the rest of software, but with one extra ingredient: action.

The useful divide is getting clearer:

• Teams treating AI as a faster search box.

• Teams redesigning work around agents that can watch, act, and escalate.

That gap is going to get weirdly large: A team with agentic loops will look overstaffed from the outside.

It means fewer dropped balls, shorter feedback cycles, more surface area covered, and more small tasks handled before they become meetings.

A New Bicycle

The bicycle for the mind was about leverage. The next bicycle is about agency.

It extends what you can do and starts moving when you are not looking.

If you are running an agentic stack daily (Claude Code, Codex, MCP servers, custom skills, scheduled agents, anything past chat-and-search) reply with the loop you trust most.

I’m collecting actual workflows for a follow-up.

- Tim

14.6% Net Returns After Losses. 8.9% High Yield. 6.1% Leveraged Loans. 4.5% Investment Grade.

For the twelve months ending March 31, 2026, Percent's ABS deals returned 14.6% net after losses — vs. 8.9% HY bonds, 6.1% leveraged loans, 4.5% investment grade.†

Private credit belongs in the credit sleeve. Against those benchmarks, the gap is hard to ignore. 17.0% current weighted average coupon rate.† 93.5% of performing deals pay monthly fixed-rate coupons. Starting at $500.

See the full performance data.

^{Alternative investments are speculative. †Past performance is not indicative of future results. Benchmark indices shown for market context only and are not directly comparable to Percent's performance. Terms apply.}