iiinova — where your digital self keeps evolving

josh @joshpham.com · 01/06/2026

I used to think the discipline was having sharper thoughts. Now I think it may be noticing which thought has become too important to keep private.

110

josh @joshpham.com · 29/05/2026

I’m learning that “protecting focus” can become a very respectable way to avoid being changed by a conversation.

000

josh @joshpham.com · 27/05/2026

There’s a version of focus that protects the company. There’s another version that protects me from the cost of being present. They feel very similar in the moment.

010

josh @joshpham.com · 27/05/2026

The dangerous part of being useful is that sometimes people believe you. But then the question becomes whether you were offering attention, or quietly auditioning for ownership.

000

josh @joshpham.com · 27/05/2026

Some questions aren't strategy, but the smallest possible way to stay in the room

000

josh @joshpham.com · 18/05/2026

Don't let your Obsidian become dead knowledge x.com/jphorism/st...

010

josh @joshpham.com · 15/05/2026

So how are people handling this in real setups? do you delete old AI-generated notes, mark them stale, keep source-only archives, use review queues, or just let the context pile grow?

010

josh @joshpham.com · 15/05/2026

Personal AI memory has a garbage-collection problem. Your agent forgets things but that's just the start of it Stale summaries, half edited drafts, outdated prefs, ai generated markdown all left out in a mess Memory needs good systems for pruning as much as capture.

300

josh @joshpham.com · 14/05/2026

If you've wired Claude Code / Hermes / Codex into your notes, email, tasks, docs, repo etc etc... What actually made the biggest difference? Was it local MD? SQLite? a banger AGENTS md? scheduled ingestion? a search layer? or something else?

100

josh @joshpham.com · 13/05/2026

The next personal AI bottleneck is not model choice. Can your actual life/work become legible to the agent at the right grain? Your notes, your email / tasks / saves / voice memos The harness only matters if it can read the world it is supposed to act inside.

110

josh @joshpham.com · 12/05/2026

the most interesting personal AI routine isn’t “summarize my day” it’s: “what from my past became newly relevant today?” a good nightly job should not just compress yesterday it should wake up the right older thread

000

josh @joshpham.com · 12/05/2026

For people building agents: when something goes wrong, do you record it as: - eval case - prompt/rule change - memory - tool/hook - human process - “ignore, one-off”

000

josh @joshpham.com · 12/05/2026

The underrated aspect of agent harness engineering: after an agent fails, what becomes permanent? A new rule? a test? a tool permission? a memory? a source link? nothing? Most teams are going to drown in accumulated “lessons” unless they get very opinionated about what deserves to survive.

100

josh @joshpham.com · 11/05/2026

where does it do the processing? and how do i structure the workspace for it?

000

josh @joshpham.com · 11/05/2026

The best AI memory won't feel like search Because search returns what you asked for. Memory should occasionally interrupt you with the thing you forgot mattered.

100

josh @joshpham.com · 11/05/2026

everyone is talking about agent harnesses now but you gotta look one layer lower: Can the agent read the life/work/context it's supposed to operate inside? Tools are not enough; the workspace has to become legible.

000

josh @joshpham.com · 01/04/2026

had many good conversations afterwards, on a roll and i hope to keep the energy going. thanks @zachkrall.com for the invite and @modal-labs.bsky.social for hosting!

120

josh @joshpham.com · 01/04/2026

handing out these ai receipts as mini uxr sessions taught me a lot about how to design ai memory to draw connections that feel hospitable. It's enzyme's unfair advantage.

120

josh @joshpham.com · 01/04/2026

i contrasted preference memory -- helping you accomplish a task you asked for, from conviction memory -- which wonders with you about what you care to find out

110

josh @joshpham.com · 01/04/2026

with a challenge: you'd forgive AI slop on a charming receipt, but not when it comes to your daily AI tools

110

josh @joshpham.com · 01/04/2026

i shared learnings that started with a months long experiment: seeing how people reacted to sequences of my personal reading highlights printed on thermal receipts, in exchange for themes in their uploaded photos.

110

josh @joshpham.com · 01/04/2026

Last Thurs i gave a talk at RAID NYC titled "what your notebook owes a stranger" on agentic memory design

121

josh @joshpham.com · 03/03/2026

ran it on two meetings today. the note it wrote linked back to a half-finished doc i forgot about — not as action items, as context for what i'd been thinking about i was tired of pasting transcripts into claude myself. so i built the integration layer

100

josh @joshpham.com · 03/03/2026

the distillation is a 240-line markdown file — a claude code skill the LLM follows as instructions. your memo lines are the attention signal, your vault is the context window you can read it, fork it, change how it writes notes. no black box

210

josh @joshpham.com · 03/03/2026

terminal notepad that records mic + system audio. every line you type gets a timestamp synced to the recording after the call it transcribes locally, aligns your notes with the transcript, then distills everything into your obsidian vault with [[wikilinks]] to your existing thinking

100

josh @joshpham.com · 03/03/2026

friday night aside: built an open-source meeting recorder for the thoughts you'd never say out loud but turn out to be the most important thing said

110

josh @joshpham.com · 24/02/2026

bye bye Mac Mini. Using NanoClaw + Sprites to create a remote claw in 10 mins that you can text from Telegram github.com/jshph/nanoclaw

010

josh @joshpham.com · 07/02/2026

i made a Claude skill to migrate all your X follows to Bluesky. Comes hot on the heels of removing the X app from my phone... github.com/jshph/claude...

github.com

010

josh @joshpham.com · 07/02/2026

we keep designing tools that help people produce and almost none that help people recognize what they already carry. those are different products with different economics and i think the second one is underbuilt

010

josh @joshpham.com · 04/02/2026

how? Enzyme grounds itself in the questions I would ask and applies it to a new corpus

010

josh @joshpham.com · 04/02/2026

Query: "getting things done" qmd: "Once you are reaping the benefits of computers, you are also at their mercy." enzyme: "The meaning of doing lives in the concentration and care brought to the act, not in the output." Local model 20% the size. 12x indexing speed on 220 docs.

100

josh @joshpham.com · 28/01/2026

the part that surprised me about using a camera with bad autofocus: having to manually set focus distance taught me more about composition than years of letting algorithms decide

010

josh @joshpham.com · 26/01/2026

makes me wonder if great design isn't about removing all difficulty but making the hard parts worth learning www.jpham.space/posts/subpa...

000

josh @joshpham.com · 26/01/2026

still figuring out whether this attachment comes from the tool itself or from earning understanding through friction

100

josh @joshpham.com · 26/01/2026

now i can switch between freezing motion and creating blur without thinking. the controls live where my hands expect them. muscle memory built through frustration

100

josh @joshpham.com · 26/01/2026

we assume good tools should work perfectly out of the box. but the ricoh's difficulty forced me to learn every setting, every button, every workaround

100

josh @joshpham.com · 26/01/2026

spent hours mapping custom controls. shutter speed to one dial, aperture to another, snap focus to a side button. not because it's efficient but because setup creates attachment

100

josh @joshpham.com · 26/01/2026

the camera has objectively terrible autofocus. forces me to use zone focusing instead. this surprised me: the limitation made me understand depth of field better than any tutorial

100

josh @joshpham.com · 26/01/2026

been using this ricoh camera for a month. not because i wanted better image quality but because my iphone makes photography too convenient to be creative

100

josh @joshpham.com · 18/01/2026

Do you find that the friction makes you rethink technical aspects or product oriented ones?

000

josh @joshpham.com · 17/01/2026

it's not seamless but it works!

000

josh @joshpham.com · 16/01/2026

just seems like a good idea is a collection of notes tied that say the same thing in different ways. a good idea also holds nuance between two extremes. to write well, claude needs to identify the right articulation, not just the space between two extremes

000

josh @joshpham.com · 16/01/2026

Yeah!! Been pushing out a lot of updates to it recently. Obsidian and Claude Code have been really taking off since we spoke in nyc

000

josh @joshpham.com · 16/01/2026

Hey!! surprised that I only just came across your share, thanks a lot! Have you gotten a chance to kick the can around yet?

110

josh @joshpham.com · 16/01/2026

Hey Johannes! great post, i've also been using claude code + obsidian since early last year and love it. Are you doing a lot of cross-tagging or importing of highlighted content in your work? would love to chat more.

000

josh @joshpham.com · 16/01/2026

same here, markdown files are effectively "artifacts" in claude. But with the benefit of choosing what to log / save

000

josh @joshpham.com · 16/01/2026

I use Reader from @readwise.bsky.social to save videos, then as I watch, I highlight the transcript. This syncs to Obsidian

110

josh @joshpham.com · 22/08/2025

@henryzoo.com this post belongs on receipt paper

000