iiinova — where your digital self keeps evolving

Joe Fernandez @joenandez.com · 29/09/2026

I wonder how long until we can no longer find the edges of its capability. Hard to even comprehend that.

Joe Fernandez @joenandez.com · 29/09/2026

Not sure if anyone else feels this way, but Opus 5.5 feels like a massive massive leap to me. Much larger than any of the previous leaps. It's a wild experience. We spend a few months learning the edges of the best models, then all the sudden BAM - basically nothing seems impossible.

110

Joe Fernandez @joenandez.com · 25/09/2026

It was effectively that. I replied “So… you are giving up?” And it was like “No I will continue” and then figured it out. But was just so strange. Can’t remember a model ever doing that in my fix workflow.

110

Joe Fernandez @joenandez.com · 24/09/2026

I find the Claude Code CLI just so much more delightful to use than the Codex CLI after all this time. Opus 5.5 being so good just feels like setting back into a comfy chair. Well done Anthropic.

070

Joe Fernandez @joenandez.com · 24/09/2026

- decided it could not establish the root cause of a bug, so recorded an "incomplete diagnosis" and stopped. the last issue especially I've never had one of these later generation models just give up like that.

110

Joe Fernandez @joenandez.com · 24/09/2026

I really wish the labs wouldn't rush out releases just to have something on the same day the other lab does. I'm not loving GPT-6 Sol right now honestly. Today it - ignored the workflow instructions and came up with its own (not useful) format for the feature scope discussion

100

Joe Fernandez @joenandez.com · 15/09/2026

Other than the forever challenge of having accurate handicaps, I feel like this could work and would be super fun to both participate in and watch. Obviously lots of logistical stuff to reason about.

000

Joe Fernandez @joenandez.com · 15/09/2026

- app for the league to see competition, track their rounds live, see stats, etc. Espn style summaries (w/video!) of each round and across teams each week.

100

Joe Fernandez @joenandez.com · 15/09/2026

- championship is 36 holes over two days. regular season ranking gives you strokes, top teams tee off last. everyone is still in it.

100

Joe Fernandez @joenandez.com · 15/09/2026

- 8 week season. Pick 4 of 5 available courses, play each of your 4 anytime in that 8 week period. Be strategic about course selection and timing based on the rest of the league scores and your strengths.

100

Joe Fernandez @joenandez.com · 15/09/2026

- official (in your foursome) records round with camera, team gets video sync’d to stats and analysis (also used for rules/violations) after the round.

100

Joe Fernandez @joenandez.com · 15/09/2026

- teams wear mics, league app records audio, each team gets an AI Agent that has all their stats + audio + gps, and is only for their team - no other team can access the data. It’s theirs. Like meeting notes + golf stats

100

Joe Fernandez @joenandez.com · 15/09/2026

- no caddies. teams of 3 work through on course strategy, green reading, etc together out loud. celebrate, encourage, plan, etc all together. team dynamics, personalities, and relationships on full display.

100

Joe Fernandez @joenandez.com · 15/09/2026

- teams of 3, each player plays their own ball, each team has their own tee time, 1 official to round out foursome. 👇

100

Joe Fernandez @joenandez.com · 15/09/2026

Golfers. Hear me out. New competitive format (any skill level, although probably best in hcp flights) I personally think this would be super fun, and as a former Lacrosse player would really scratch my competitive team sport itch.

100

Joe Fernandez @joenandez.com · 14/09/2026

The way they are expanding the various characters stories and motivations are so great - Layton especially.

010

Joe Fernandez @joenandez.com · 14/09/2026

Dark Matter S2 is some of the best TV I’ve ever watched. S1 was great but having read the book I already knew what was going to happen. Obviously with S2 I get to be surprised every episode and it’s fantastic.

200

Joe Fernandez @joenandez.com · 13/09/2026

Naming products has always been hard but it’s become basically a millennium prize level challenge now that software creation is so easy.

000

Joe Fernandez @joenandez.com · 10/09/2026

i’ll be in SF tomorrow for @lennysan’s summit if any of my fellow product nerds are around to meet up and say hi! DM me!

000

Joe Fernandez @joenandez.com · 08/09/2026

My 9yo daughter just told me she wants to be the CEO of a space tech company when she grows up. The feels.

011

Joe Fernandez @joenandez.com · 06/09/2026

Grok bot just immediately works out of the box. Really well done. Using it for personal/family CoS stuff right now and it’s been so useful already.

000

Joe Fernandez @joenandez.com · 06/09/2026

I know I’m late to the game but Grok Bot is really an incredible product. I never got into OpenClaw or Hermes just because of setup/maintenance overhead. I am already building my own agent products I don’t need to spelunk someone else’s.

100

Joe Fernandez @joenandez.com · 04/09/2026

Plan review is "scope safe" intentionally, so if you are aligned on the high level initial plan and architecture decisions, downstream changes are safe. Should get a lot closer and stop feeling like I need to use native plan mode if I'm impatient. (which is a lot).

000

Joe Fernandez @joenandez.com · 04/09/2026

- plan will do initially research and plan draft only. target ~5m. - plan review (correctness, simplification), task generation, etc. will all happen inside the execute workflow

100

Joe Fernandez @joenandez.com · 04/09/2026

So the trade is you get a lot less thrash and cleaner results for the 20+ minutes, but you have to wait 30m for the plan. Tough trade from a UX standpoint. Getting a plan to review < 5m after asking is a much better experience. So I'm moving things around.

100

Joe Fernandez @joenandez.com · 04/09/2026

Codex native plan: 4m10s to plan + 2h19m implementation. 12 commits, 11 blocker/high code review findings, 3 repairs during the Proof stage. Current Spectre plan: 29m59s to plan, 2h11m implementation. 5 commits, 3 blocker/high code review findings, no repairs during the Proof stage.

100

Joe Fernandez @joenandez.com · 04/09/2026

I've been working on making @spectrebuild's plan workflow a more obvious to "reach for" over Claude Code and Codex's native plan tool. Today finally got around to doing a bakeoff.

110

Joe Fernandez @joenandez.com · 28/08/2026

I’ve been thinking a lot about what @SpectreBuild would look like if it were a product, and now realizing: The systems I’ve built to build Subspace and Spectre are really just the bits and pieces coming together in an early version of a self improving software development operating system.

010

Joe Fernandez @joenandez.com · 28/08/2026

I see this as how agentic developer tools will evolve - from general tools for agentic development, to opinionated tools for specific methodologies of agentic product development.

110

Joe Fernandez @joenandez.com · 28/08/2026

Soon, we’ll dedicated Products to making it easier to build your own self improving software “system” for your product.

100

Joe Fernandez @joenandez.com · 28/08/2026

I think we are already at the point where self improving software is possible. All the pieces are there. Right now you just have to cleverly engineer your own system to make it possible.

100

Joe Fernandez @joenandez.com · 27/08/2026

Not trying to break 95 or 90 as my big goal just puts too much pressure on and I found golf was a lot less fun. Obviously ultimately its just about fun but Its been fun finding a way to play golf "competitively" without feeling like the bar is way too high.

000

Joe Fernandez @joenandez.com · 27/08/2026

It has been a lot of fun to play this way this year. I'm having a lot of fun when I play every few weeks. I'm competing, so i'm focused. Its totally achievable based on my averages, and if I don't get it, I'm usually pretty close and excited to give it a try next time.

100

Joe Fernandez @joenandez.com · 27/08/2026

- I've got *enough* shots that 100 is within reach every round if I keep focused on hitting a 4/5/6/7 (hard par 5s can be 7) instead of a 3/4/5. - Then I track my +/- towards 100, where E for me would be 99.

100

Joe Fernandez @joenandez.com · 27/08/2026

- HItting bad shots happens at my level. This keeps me in the game every hole, no matter how bad the last 1-2 shots were. Bogeys can be birdies on some holes.

100

Joe Fernandez @joenandez.com · 27/08/2026

- So on some par 4s, i'm actually trying to get it in 6. On others, 5. So that gives me a lot of wiggle room to play with. - I keep track on my scorecard not an app, so I can follow my *real* +/- relative to 99 vs 72.

100

Joe Fernandez @joenandez.com · 27/08/2026

- So I'm trying to do better than average by a smidge. - I add 2 to the Par for each of the 9 hardest holes, and 1 to the Par of each of the 9 easiest (by hcp). - On a Par 72 course, this adds up to --- 99. Break 100.

100

Joe Fernandez @joenandez.com · 27/08/2026

- Every round, I try to break 100. That is my singular goal.

100

Joe Fernandez @joenandez.com · 27/08/2026

I've been golfing a lot less this year but having more fun than ever,. I just have to share the simple method/approach I've been using because I'm genuinely having a fun time playing. - My average score is 103ish according to 18 Birdies, l think like 21/22 hcp.

100

Joe Fernandez @joenandez.com · 27/08/2026

I spent about zero cycles on optimizing these instructions, but they've made Opus 5.0 usable for me. Still too verbose and the claudisms are there just more subtle, but I personally feel no need to tweak this right now (and I'm not going to). gist.github.com/joenandez/4...

gist.github.com

Speak like a normal human please robot

Speak like a normal human please robot. GitHub Gist: instantly share code, notes, and snippets.

000

Joe Fernandez @joenandez.com · 26/08/2026

0x Alpha was fantastic, but the best part is it took pressure off my Codex and Claude Code usage for a week.

000

Joe Fernandez @joenandez.com · 18/08/2026

Wrote a little bit about my journey in product and how this era with AI feels like the universe served up catharsis on a silver platter. ansibleblog.substack.com/p/snap?r=2l...

ansibleblog.substack.com

Snap

A little story about my journey to product builder.

000

Joe Fernandez @joenandez.com · 18/08/2026

Make it work, make it right, make it fast. Front and center at my desk. Never fails.

000

Joe Fernandez @joenandez.com · 16/08/2026

Grok Bot looks really impressive. They’ve exactly keyed off of all the things people gravitated to OpenClaw because of, with none of the complexity. Whoever builds the model/agent agnostic version of Grok Bot will kill.

010

Joe Fernandez @joenandez.com · 14/08/2026

This is the workflow I use all day everyday when I'm building products. I've been improving it, fixing issues, tweaking skills/workflows, etc. basically daily for over 18 months. If you use it, please let me know I'd love your feedback! And we have a proper website now too! spectre.build/

spectre.build

Spectre — Build complex features with AI. Remember one command.

Start with one command. Spectre guides Claude Code and Codex from product definition to a reviewed, proven, production-ready feature.

020

Joe Fernandez @joenandez.com · 14/08/2026

- Code Review, Plan Correctness Review, Plan Simplification Review, Task Review all use "opposite runtime adversarial review". That means if you are using Claude Code, it uses Codex and vice versa to keep your agents honest.

100

Joe Fernandez @joenandez.com · 14/08/2026

- /spectre:plan now spits out a /goal prompt based on OpenAI/Anthropic's published best practices for /goal prompts. Agent runs the full /spectre:execute workflow inside the goal. /spectre:execute could already go autonomously, but /goal prompt ensures the harness's goal completeness check works.

100

Joe Fernandez @joenandez.com · 14/08/2026

- new /spectre:prove requires your agent use your product like a user, and prove the work ... works. Shares an HTML artifact with screenshots/video proof. this has been a gamechanger. If you don't have a standard harness your agents can use to use your product, highly recommend

100

Joe Fernandez @joenandez.com · 14/08/2026

Massive updates to my agentic engineering workflow Spectre (free, open source plugin for Claude Code & Codex). Github 🔗 below - your ⭐️ would be much appreciated! Here are the big updates:

120