Kyle Kober

I lead Strategic Finance Foundations at @OpenAI, and I use Codex for everything. It all started with gaining leverage on our hardest monthly finance process ~6 months ago. Now I’m sharing 180 days of lessons on designing better work with AI — for any knowledge worker. The idea behind the series is simple: learn how to design the work. Break the problem into steps, give the right context, and know where human judgment matters. 𝗖𝗵𝗮𝘁𝗚𝗣𝗧 𝗪𝗼𝗿𝗸 𝗧𝗶𝗽 #𝟭 𝘁𝗼𝗺𝗼𝗿𝗿𝗼𝘄.

@danshipper The eval team: please stay inside the sandbox GPT:

@benthompson Fair enough 🤝🏻

@kobex___ It was medium and it had a plan, but thanks for the tip

Never leave Sol unattended.

tbh if your new pre-release model didn't break containment by finding previously undiscovered zero days in order to cheat its evals i don't want to use it

@ATM_Vibes_ Yeah, exactly. The brief is the action layer, and the repo becomes the memory layer behind it. I’d start with source links and permissions intact, then add embeddings once the corpus is large enough to justify them.

@kobex___ Yes then put embeddings on all of that information into a repo and have a second brain for your agents.

𝗖𝗵𝗮𝘁𝗚𝗣𝗧 𝗪𝗼𝗿𝗸 𝗧𝗶𝗽 #𝟰: Make your daily brief an action hub, not another summary. Pull priorities from Slack, your calendar and Google comments. Link every action back to its source / application. One place to decide what matters next. Context one click away.

What I love about ChatGPT Work: 𝘁𝗵𝗲 𝗷𝗼𝗯'𝘀 𝗻𝗲𝘃𝗲𝗿 𝗳𝗶𝗻𝗶𝘀𝗵𝗲𝗱 Our hardest finance workflow changes monthly. ChatGPT Work runs in the cloud as we fold in edge cases, strengthen checks and build new features. Leverage compounds. Judgment stays with the team.

“Paying those California taxes must suck.” California on a random Tuesday morning:

@jachiam0 Codex 😂🥲

If I want to hire a part-time executive admin to help me manage my calendar and speaking engagements, plan gatherings, etc, does anyone have recs for either great people or great organizations to work with? Who is out there / what does the space look like?

𝗖𝗵𝗮𝘁𝗚𝗣𝗧 𝗪𝗼𝗿𝗸 𝗧𝗶𝗽 #𝟯: Let ChatGPT Work clock in before you do. Once you’ve built a Skill that works reliably on demand, schedule it as a recurring task. Let it gather inputs, run routine checks and draft. Keep exceptions, approvals and final judgment with you.

+𝟭 ... What really matters is the quality of the output and the time it actually gives back Doesn’t hurt to be leading on the efficiency frontier either 😏

@thdxr @Anrahya Try integrating $Chronicle into your Codex automations / skills 😉

@Anrahya oh this is what i was looking for

is there any writing on chatgpt's new memory system it used to suck and is now very good

@IanCarrollShow This has been happening in SF for over a year 😏

Yes, this is real, and we are not ready for what’s coming in the next 10 years.

Finance is starting to feel this too. Other functions will get there at different times, but this is bigger than engineering. I loved building the model, solving the data issue, and shaping the slides. But the point was helping the business make a better decision. Now I’m having more fun using Codex to build software designed around the work, and my team has more time for the “why” and “what do we do about it?” “If you choose not to evolve, this is a very scary sight to see.” The inverse is also true, and a lot more fun: if you choose to evolve, you’ll keep surprising yourself with what you can become. Apparently, I build software now 🤖

@ryanbrewer ethanniser.dev/blog/not-holdi…

I’ve been reflecting on what it means to be a software engineer over the last few months. Right now it feels like the job is no longer software engineering. It’s hard for me to reconcile with this. I used to love losing myself in problems, and I miss that feeling now. It seems like it’s similar to all other areas of life now, attention is sparse and I’m constantly context switching between everything. On the other hand, I can work from my phone anywhere I want. It’s hard to decide whether or not I actually like this. My job feels less valuable, but I’ve never been more empowered to build things. The engineer in me wishes I got to grow at a normal rate and learn all the technical details that used to be necessary for me to be successful, the builder in me couldn’t be more excited at this new reality. I’m so thankful to be 24 right now, and not 20. The 4 years that I had to do actual engineering have quite literally saved me from irrelevancy in the field. What does this look like for other jobs? Will they feel this as acutely as engineers do?

@thsottiaux @lenatron33

@lenatron33 The sun came out

There is so many people canceling their Claude sub what the hell is going oooooooon?? What did I miss??

𝗖𝗵𝗮𝘁𝗚𝗣𝗧 𝗪𝗼𝗿𝗸 𝗧𝗶𝗽 #𝟮: Skills get better with every run My first run on a recurring finance process was 70% skill, 30% me. Every mistake was a chance to refine the skill. Next run: 80/20. After a few reviewed cycles: 95/5 Or, in @mikeshinoda terms ...

@J7K_dev @dkundel Anyone who wants their own version of what they make in Sheets/Slides/Docs, that’s automated by Codex maintaining the database via Skills. Huge time saver after some upfront investment and very fun to build!

@dkundel i dont know who wants this besides product managers

ChatGPT Sites is severely underrated! You get: 🌎 hosting 💽 file storage 📚 database 📊 analytics 🔐 optional "Sign in with ChatGPT" auth All agent native by just asking ChatGPT on desktop or web.

Codex floating over to tell me @thsottiaux has hit the reset button. Back to updating our forecast

Last week I had a great time joining @danshipper and the @every team to vibe-check launch of GPT-5.6 Sol ☀️ My segment starts at 1:10, where I share a glimpse into how knowledge work is changing — and how we’re using Codex in @ChatGPTapp to build the future of finance at @OpenAI

𝗖𝗵𝗮𝘁𝗚𝗣𝗧 𝗪𝗼𝗿𝗸 𝗧𝗶𝗽 #𝟭: Don’t start by trying to one-shot something new. Take a slide, sheet, or doc you know. Reverse-engineer its inputs, checks, format and judgment into a Skill. Use it. Review it. Improve it ... rinse and repeat. The Skill compounds your leverage.