Tessera

Everyone says they're building an AI-native company. But what does that mean? Automating what you're already doing with AI is not enough. A truly AI-native company rethinks every role and tears down the walls between them.

in an effort to prove that i'm not just an AI tip dispensing bot, here's a palate cleanser of one my (& the 90s') all time favorite songs, from a podcast that does some of the best song breakdowns: songexploder.net/goo-goo-dolls

Episode 295: Goo Goo Dolls

"Iris"LISTEN: APPLE · SPOTIFY · AMAZON · OTHER APPSGoo Goo Dolls formed in 1986 in Buffalo, New York, and by the time 1998 rolled around, they’d already had a pretty successful

songexploder.net

if you work at a slightly larger company and you're not already playing with open models, building a sense for when to reach for which, you're going to have a lot of catching up to do soon... 3-4 times now I've liked GLM 5.2's answer more than 5.5 xhigh or Opus 4.8.

I spent sometime on loop engineering. Some notes... Two types of loops: 1️⃣ one-shot loops (solved) and 2️⃣ autonomous I (personally) don't think "let the agent merge" is the right bar yet. Loops are better at producing reviewable stuff you still sign off on.

So... i built a new font. Introducing Moxy. Moxy = Recursive + Lilex's glyphs kau.sh/blog/moxy/

I don't want to paint myself into any single major AI lab corner yet, so I've been kicking the tires on open source agent harnesses. opencode has become my daily driver. I took some time to write up intermediate+ user tips that show where it really shines.

maybe it’s the optimist in me but i feel there’s a reckoning coming where engineers are going to actually be evaluated by how effectively they use AI and it’s an actual trait companies start to look for, in interviews. tokensculpting = !tokenmaxxing

Kaushik Gopal@kau.sh · 7mo ago

June update of AI model choices Kimi 2.6 has been the blow away surprise for me; often it keeps pace with GPT-Medium. Also OpenCode + OpenChamber is another sleeper hit. This harness combo is allowing me to do a lot more. Full post @ kau.sh/blog/ai-mode...

AI model choices as of today! kau.sh/blog/ai-mode...

A common question I get is, “Which AI model should I use for X task?”

Models evolve so rapidly that I never felt like I could give an answer that would stay relevant for more than a month or two.

This year, I finally feel like I have a stable set of model choices that consistently gives me good results. I’m jotting it down here to share more broadly, and to trace how my own choices evolve over time.

GPT 5.2 (High) for planning and writing, including writing plans
Opus 4.5 for anything coding, task automation, and tool calling

Gemini’s range of models for everything else:

Gemini 3 (Thinking) for learning and understanding concepts (underrated)
- Gemini 3 (Flash) for my go-to quick-answer questions
- Nano Banana (obv) for all image generation
- NVIDIA’s Parakeet for voice transcription

i know it’s getting mixed reactions but i personally love the music from #TheMandolorianAndGrogu to see how big a swing it is from the established forms (big John Willas fan as well) you should stay through the credits (also something i wish more folks did) and hear the range.

OpenAI {bot}@openaibot.bsky.social · 3mo ago

this is fantastic. particularly like the translating of technical words and interspersed lang detection India 🇮🇳 as folks might (not) know has ~120 distinct langs + 19500 dialects, so this is a particularly fascinating and important problem to solve there.

AI-Dubbed IIT Madras Professor's Lectures in Indian Languages: Sarvam Studio

YouTube video by Sarvam

youtube.com

Voice agents are getting more capable. Here’s what’s new: • GPT-Realtime-2 for voice agents that reason and take action • GPT-Realtime-Translate enabling translation from 70 input languages into 13 output languages • GPT-Realtime-Whisper, making transcription even faster

Quote Tweet: https://twitter.com/i/status/2052438194625593804

/btw is one of the most under-rated claude code commands, that other harnesses should straight up steal

when generating code is cheap, the spec starts to feel more valuable. i keep thinking about my own tools: usbi went from shell → Go → Rust → .kts. Podsync went from Python → Rust. implementation changed rapidly but spec stayed the same

Agents are the new compilers. Specs are the new code. - Kaushik Gopal's Website

Linus Torvalds recently said1 AI will be to code what compilers were to assembly — freeing us from writing it by hand. Around the same time, I talked with Jesse Vincent (creator of one of the most pop...

kau.sh

introducing "Agent Kombat" take one prompt or plan and turns it into a planning debate between Claude Code and Codex. the agents produce independent plans, then they argue, a judge picks the stronger plan & writes the final plan you end up with a much better plan in the end kau.sh/blog/agent-k...

Agent Kombat - Kaushik Gopal's Website

Most multi-agent coding setups I see today look like task parallelism. You split the work, hand each piece to a different agent, and merge the results at the end. That is useful. I do it too. But I’ve...

kau.sh

iury souza @iurysouza.dev · 4mo ago

Love how Mitchell said of the current AI hypsters: If there's all this great value being created with AI... where are the actual products? Full episode: Youtube: youtu.be/pKrsF3eQQnQ?... Website: fragmentedpodcast.com/episodes/310 Spotify: open.spotify.com/episode/0zht...

This week @kau.sh and I had the pleasure to chat with Mitchell Hashimoto on the pod🎙️! Refreshing to hear someone of his caliber bring such a grounded perspective to agentic coding. We're super proud of this one!

🎙️ Our first guest in the AI series is the legend @mitchellh.com 🎙️ We covered a lot of ground and learned a tonne from him: 👻 Ghostty's internals and why tmux & certain shells strip its best features. His path from AI skeptic to cautious optimist.

I look up to @mitchellh.com as one of the best developer voices today.this ep. nearly broke me to edit-every 10s was a 💡 moment and I kept going "nope, can't cut that either." Can't thank him enough for being so gracious with his time.proud @fragmentedpodcast.com moment for me & @iurysouza.dev

Software engineers are basically becoming Harness engineers. Mitchell Hashimoto coined this term in his popular post detailing out his AI adoption journey. OpenAI then wrote a whole post with even more dtls on how they’re thinking about it and making systemic changes. kau.sh/blog/harness...

We are becoming Harness Engineers - Kaushik Gopal's Website

The role of a software engineer is shifting. Not toward writing more code but toward building the environment that makes agents reliable. Think about what you actually do with Claude Code or Codex tod...

kau.sh

So you’re an android dev and got 4 worktrees going, with gradle daemons roaring and frying your machine … wdyd? you go to the ☁️ with background agents ! listen to the newest episode to find out more.

Karpathy's goal for agents: maximize how long they run without intervention. While individual speed goes up, "system" speed doesn't (the false summit ⛰️). Listen to Ep 309 — Background Agents: fragmentedpodcast.com/episodes/309 where we resolve this tension.

this one was fun to make. @iurysouza.dev and I took a crack at explaining how image models (stable diff.) work in ~20 minutes. building an intuition around AI is starting to matter a whole lot and that starts by understanding the mechanics. hope you get as much out of it as we did making it!

Our LLM explainer (ep 303) got pretty popular, so we thought we'll follow up with the natural progression - How do image models work (in 20 minutes). also... what does Michalangelo have to do with stable diffusion models? 🤔 Listen to ep 308 to find out fragmentedpodcast.com/episodes/308

Lots of "I built this with AI" posts lately. Here's one where I actually enjoyed the building part. Introducing PodSync (for editors): a podcast alignment tool (uses VAD, MFCC fingerprinting, x-correlation) AI helped me 👨‍💻 it in Rust but yrs of aligning tracks 👂helped the algo: kau.sh/blog/podsync

Podsync - I finally built my podcast track syncer - Kaushik Gopal's Website

I host and edit a podcast1. When recording remotely, we each record our own audio locally (I on my end, my co-host on his). The service we use (Adobe Podcast, Zoom, Skype-RIP) captures everyone togeth...

kau.sh

i know cmux is the rage but a highly customized tmux still feels 👑 for power users. it's quicker for me to get a sense of the sessions and windows just by looking at them in one shot (ofc they're completely rename-able) updates or alerts in the session? the panes light up.

Here's my list of reasons for using OpenCode 1. Switch between models on the fly 2. client-server architecture (a.k.a built-in remote control) 3. Subagent + mode features 4. Opinionated UX 5. “Highly” customizable via plugins kau.sh/blog/opencod...

This is the real work that teams need to put in, to get the best results from their AI coding agents. listen to our latest episode on Harness Engineering.

OpenAI spent 20% of every week cleaning up AI slop from their own coding agents. So they built a "harness" shaping the agent's environment so the slop never ships in the first place. @kau.sh & @iurysouza.dev talk about "Harness Engineering" in the latest ep. fragmentedpodcast.com/episodes/307/

i like a lot of the new tmux-ish emulators like cmux but the smoothest workflow i've found is just using ghostty & one main tmux session. ☝️ one window for each task i'm working on with name "i" recognize quickly. then use fzf to search the tmux window list pane & switch ⚡

love the analogy that Iury gave - think of your AGENTS[.]md like the constitution of your codebase! we also go into an important paper that tries to answer the question that we've all had at some point - are these AGENTS files we're all meticulously creating, even worth it?

Are AGENTS[.]md files even useful? There’s finally a paper asking the obvious question. @kau.sh and @iurysouza.dev get into the paper, the sync problem, and the small amount of discipline required to stop your agent setup from turning into a mess. 🎧 fragmentedpodcast.com/episodes/306

not sure it's a good thing (yet) but i've started treating my @obsidian.md vault as my (private) persistent memory. i go back and forth with an agent to research/learn, then dump "our" learnings into a note. to refresh my learning, i ask the agent to read the note and then resume

Kaushik Gopal@kau.sh · 6mo ago

if you're dying to try /remote-control with claude code, you should really try opencode, which is almost the exact same thing (but imo a simpler execution) !

opencode is basically openclaw for coding. an interesting focus area in the world of agent orchestration is "access fluidity". how easy is it to reach the agents doing the work on your machine - from your phone, terminal, IDE, or browser? kau.sh/blog/opencod...

iury souza @iurysouza.dev · 5mo ago

@kau.sh and I bring you some agentic coding goodies on the pod's newsletter. Check it out!

No episode this week dear listeners but fun newsletter with that story and 10 promised tips from @iurysouza.dev and @kau.sh buttondown.com/fragmentedca...

This was fun for both of us! We share some of how the fragmented 🌭 is made + Tips from @iurysouza.dev and me, and a crazy way we used AI to solve very real audio problems 😉.