Co-Founder @specstoryai. Organize D.C. AI tinkerers. Use https://xitter.gq/t.co/kg5JldHaiU, a calm agent multiplexer with familiar IDE features, for macOS.
Washington D.C.
Joined October 2015
- Tweets1.9K
- Following470
- Followers1.1K
- Likes1.4K
Pinned Tweet
hAvE yOu hEaRD aBoUt: omniagent, paseo, orca, t3code, herdr, cmux, etc etc?
ofc but I drive my with agents sessions with a cat.
👉tortie.sh/
a calm agent multiplexer that feels like vscode, doesn't make you learn tmux and does everything it can to save your work
Highly recommend you start doing this. Here is an example of how long an org and billing spike took in precise "agent hours" (new estimation unit to replace eng weeks?)
Curious about how to get started? Read: gregceccarelli.com/writing/h…
Using claude code workflows extensively with access to session traces and message timestamps has an unexpected fringe benefit: am able to produce very precise estimates of total wall clock time required to spec, build and fully verify features.
Before: claude was like "Oh this plan will take 2 to 3 engineering weeks" and we all knew that was wrong.
Now: claude is like it will take 7 hours of agent build and integrate implementation time, 5 - 6 hours to verify, fix and reverify and 45 minutes to release to prod via your process based on the last 60 phases you ran and my calculation of spec complexity, files touch, verification scenarios to run through for this change and your CI/CD timings.
Its been accurate within +/- 10 pp over the last few weeks.
You would not even believe the amount of credentials / keys (aws, gcp, azure, saas vendors, oai, ant, etc) exposed to models via agents on a daily basis in agent session logs.
I don't even think you can even fathom the volume of exposure.
The Trezor Safe 7 is a beautiful piece of hardware with software that is an equal complement.
Upgrading from the Trezor One after years of use was well worth it.
Really an incredible product.
Expert level baiting
despite the unfortunate timing, we take disclosures seriously at @typesafeai
at approximately 6:49PM this Friday evening, we became aware of a massive unauthorized leak of ALL customer noul outputs
in effort to maximize transparency, here is the link: gist.github.com/diogo149/810…
Greg Ceccarelli retweeted
Since developers aren’t coding anymore, their new job is to rotate secrets every single fucking day
Greg Ceccarelli retweeted
>junior dev at a mid startup
>wednesday morning
>just want to push a minor bugfix to our php app today
>open laptop, check slack
>company group chat is an absolute warzone
>CTO typing in all caps about php2shell
>fucking php
>whatever, server hasn't blown up yet, I just need to get my code in
>open gitlab to merge the PR
>forced emergency logout because of a critical gitlab security patch release
>fucking gitlab
>pipeline fails because of some weird regex compiler bug
>spend 40 minutes fighting CI/CD
>i can just use AI to help debug
>fire up opencode
>api keys nuked because they were breached
>fucking opencode
>i’ll just get fresh LLM api keys from our corp proxy, LiteLLM
>it won’t connect? It’s down?
>look up LiteLLM news: supply chain compromise
>fucking litellm
>cold sweat forming
>slack alert: db connection error
>check our database connection
>we host on TiDB Cloud
>they just announced a security breach, unauthorized access via stolen service account credentials, data copied
>fucking tidb cloud
>full panic attack incoming
>figure fuck it, just ship the container to Railway and go home
>open Railway dashboard
>bright red banner: "Security Disclosure: Rotate all environment variables immediately. All builds paused."
>fucking railway
>try to open the app URL in browser to see if anything is even working
>NET::ERR_CERT_AUTHORITY_INVALID
>eMudhra intermediate certificates compromised
>4
>instantly untrusted by every major browser
>fucking emudhra
>stare blankly at mac screen for 10min
>random pop-up
>restarting in 10min to fix a critical vulnerability
>fucking macos
>close laptop
>leave office
>never come back
By popular request, field notes with worked examples on exactly how I use estimate future agent work from logs to escape the wrongly calibrated, default and ungrounded "yeah this spec and work will take 2 to 3 engineering weeks" we all dread.
Can skip to a straightforward getting started section!
gregceccarelli.com/writing/h…
Using claude code workflows extensively with access to session traces and message timestamps has an unexpected fringe benefit: am able to produce very precise estimates of total wall clock time required to spec, build and fully verify features.
Before: claude was like "Oh this plan will take 2 to 3 engineering weeks" and we all knew that was wrong.
Now: claude is like it will take 7 hours of agent build and integrate implementation time, 5 - 6 hours to verify, fix and reverify and 45 minutes to release to prod via your process based on the last 60 phases you ran and my calculation of spec complexity, files touch, verification scenarios to run through for this change and your CI/CD timings.
Its been accurate within +/- 10 pp over the last few weeks.
Hot take: long running agents (e.g. using Opus 5.5 w/ ultracode) are so adept at being orchestrated that building software with LLMs has become pedestrian.
~12 months ago using Opus 4.5 still required substantial steering but you really did _feel_ amplified. I remember being absolutely engrossed in the work and looking forward to figuring out what to do next -- whether it was research, planning & speccing, reviewing implementations, figuring out how to translate more ideas into action, etc. All the while combatting (AI Vampirism).
~6 months ago when Opus 4.8 and "/goals" were introduced I was enamored with the possibility of much longer running sessions and figuring out how to construct the right upfront verification requirements. There was the challenge of writing good goals and getting them to run autonomously.
Now once you have a reasonably articulate definition of what you want and a defined process for how to research and build, you barely have to _do anything_ and I can sum up the residual of what was once still a creative process after writing code diminished as just "executive review".
Case in point: over the course of the last 30 days I've pushed over 105B tokens through agentic workflows running 24/7 to build tortie.sh and specstory.com/runstory.
Most days amount to substantial accretive change in each which is exciting but what I'm personally doing is light management if you can even call it that.
I spend a few hours at the start of a week designing the next set of backlogged phases of work to complete and just unblock the agents that fully spec -> build -> integrate -> verify -> fix -> reverify each phase until I end up using what comes out the tail end.
Ofc I can actually go touch more grass now: nice! But manipulating digital bits into existence with agents has started to become as dull and soulless as managing large teams of humans eventually did.
I didn't expect to hit this wall within 2 years and I'd assume others will or do already feel the same.
Imagine by contrast being Accenture who yesterday rolled out $29/mo Claude spend limits to their 800K employees. Which annualizes assuming full consumption to $280M a year in spend.
Some if not most companies have zero understanding of power laws.
Using claude code workflows extensively with access to session traces and message timestamps has an unexpected fringe benefit: am able to produce very precise estimates of total wall clock time required to spec, build and fully verify features.
Before: claude was like "Oh this plan will take 2 to 3 engineering weeks" and we all knew that was wrong.
Now: claude is like it will take 7 hours of agent build and integrate implementation time, 5 - 6 hours to verify, fix and reverify and 45 minutes to release to prod via your process based on the last 60 phases you ran and my calculation of spec complexity, files touch, verification scenarios to run through for this change and your CI/CD timings.
Its been accurate within +/- 10 pp over the last few weeks.
Tortie for Mac 0.111.1 is out, and w/ it Tortie for iPhone is in public beta via TestFlight.
Tortie runs your coding agents (Claude Code, Codex, Cursor, Pi, Grok and many more) in one calm window, and now you can reach them from your phone.
Your phone shows your real terminal, not a chat-style copy of it. You see the same screen your Mac shows, and you can scroll back, type, or answer a question with a tap.
No account and no server. Your phone talks only to your Mac, through your own Tailscale Funnel on that Mac. You pair once by scanning a QR code.
Free and open source. Mac: tortie.sh. iPhone beta: tortie.sh/iphone
hAvE yOu hEaRD aBoUt: omniagent, paseo, orca, t3code, herdr, cmux, etc etc?
ofc but I drive my with agents sessions with a cat.
👉tortie.sh/
a calm agent multiplexer that feels like vscode, doesn't make you learn tmux and does everything it can to save your work
The irony of gridcraft being hosted on GitHub is palpable github.com/storytold/gridcra…
But it’s not going to change the behavior of 99.9% of excel users.
As I predicted just a few days ago. I said Microsoft and Excel were going to be next.
Brandon Thomas isn't just trying to provide compatible replacements for these tools. He's attempting to break the entire enshittified business model that produced them.
More power to him!
tomshardware.com/software/mi…
I don’t think it’s clear to anyone what the catalyst will be to change this distribution.
It may be that the scale of ambition to take advantage of the agentic leverage now possible is relatively condensed. Understanding what’s possible to achieve and having the ability to drive agents to accomplish what the frontier understands are its capabilities might stay very unevenly distributed.
I’m not sure this is good.
Dusting off this tweet from April because this gap in shared understanding of LLM capability is *widening*. It's now less "two groups of people speaking past each other" and more a sharp funnel.
- Napkin math somewhere around 6B people (~75% of the population) have barely come in contact with LLMs at all.
- Around 1-2B (~20%) are casual and infrequent users of free-tier ChatGPT-like products. This group sees derpy chatbots and treats them a bit like a better Google search, a writing aid, or etc. Maybe an agent tries to book you a flight. I have non-tech friends who (reasonably, imo) say they have not much use for it at all. Even many professionals outside of math&code are in this tier. For example, execs and many other functions spend a lot of their time talking to other people, so while they understand what is happening intellectually, it is still second-hand and a bit abstract.
- Now we get to professional use of frontier-grade LLMs in math&code. Somewhere around 20M people (0.2%) see first-hand that large, complex projects that used to take them weeks/months can now be completed by agents with a prompt. Building apps, copying apps, translating apps, decompiling apps from binaries... This has all happened very quickly and recently - less than 1 year ago, I was writing code manually by hand, typing memorized computer code commands into a code editor character by character, occasionally pressing Tab to autocomplete a little chunk of code.
- And finally we get to the ~5,000 people (~0.00006%) with access to frontier-grade systems internally. The external world has seen the preview. It looks like swarms of thousands of agents collaborating over weeks on software mega projects: minting zero days, running cyber attacks and defenses at machine speeds, discovering new science, advancing the frontier of mathematics. Things that would have taken top professionals in the industry years of work. Meanwhile, human review and comprehension are starting to fall behind. For example, people are still involved in the "archeology" of the OpenAI-HF incident from many months ago. Mathematicians may be poring over the 722 manuscripts on frontier mathematics for a while.
The funnel is driven by a combination of factors:
1. The impact scales with ambition, problem size, and horizon. A question with a paragraph answer barely stresses the system. You need a reservoir of big, difficult problems that you really care about. This aspect drives the consumer / professional dimension of the funnel.
2. The jaggedness of the system (which I have written about a lot separately). Capability peaks in domains that are digital, verifiable and economically valuable. This is because LLM capability emerges from reinforcement learning on verifiable rewards on a curated environment mixture driven by revenue potential. This aspect primarily drives the area (e.g. math&code) dimension of the funnel.
3. Access. Free-tier, paid-tier, internal.
So this is the weirdness of the moment. The general public has mostly not interacted with these systems. When they have, it looks like a derpy chatbot. The majority of professionals still see only a modest uplift. And a small sliver of professionals are experiencing the vertigo of the curve going vertical. And it is all happening at the same time.
Greg Ceccarelli retweeted
Dusting off this tweet from April because this gap in shared understanding of LLM capability is *widening*. It's now less "two groups of people speaking past each other" and more a sharp funnel.
- Napkin math somewhere around 6B people (~75% of the population) have barely come in contact with LLMs at all.
- Around 1-2B (~20%) are casual and infrequent users of free-tier ChatGPT-like products. This group sees derpy chatbots and treats them a bit like a better Google search, a writing aid, or etc. Maybe an agent tries to book you a flight. I have non-tech friends who (reasonably, imo) say they have not much use for it at all. Even many professionals outside of math&code are in this tier. For example, execs and many other functions spend a lot of their time talking to other people, so while they understand what is happening intellectually, it is still second-hand and a bit abstract.
- Now we get to professional use of frontier-grade LLMs in math&code. Somewhere around 20M people (0.2%) see first-hand that large, complex projects that used to take them weeks/months can now be completed by agents with a prompt. Building apps, copying apps, translating apps, decompiling apps from binaries... This has all happened very quickly and recently - less than 1 year ago, I was writing code manually by hand, typing memorized computer code commands into a code editor character by character, occasionally pressing Tab to autocomplete a little chunk of code.
- And finally we get to the ~5,000 people (~0.00006%) with access to frontier-grade systems internally. The external world has seen the preview. It looks like swarms of thousands of agents collaborating over weeks on software mega projects: minting zero days, running cyber attacks and defenses at machine speeds, discovering new science, advancing the frontier of mathematics. Things that would have taken top professionals in the industry years of work. Meanwhile, human review and comprehension are starting to fall behind. For example, people are still involved in the "archeology" of the OpenAI-HF incident from many months ago. Mathematicians may be poring over the 722 manuscripts on frontier mathematics for a while.
The funnel is driven by a combination of factors:
1. The impact scales with ambition, problem size, and horizon. A question with a paragraph answer barely stresses the system. You need a reservoir of big, difficult problems that you really care about. This aspect drives the consumer / professional dimension of the funnel.
2. The jaggedness of the system (which I have written about a lot separately). Capability peaks in domains that are digital, verifiable and economically valuable. This is because LLM capability emerges from reinforcement learning on verifiable rewards on a curated environment mixture driven by revenue potential. This aspect primarily drives the area (e.g. math&code) dimension of the funnel.
3. Access. Free-tier, paid-tier, internal.
So this is the weirdness of the moment. The general public has mostly not interacted with these systems. When they have, it looks like a derpy chatbot. The majority of professionals still see only a modest uplift. And a small sliver of professionals are experiencing the vertigo of the curve going vertical. And it is all happening at the same time.
Judging by my tl there is a growing gap in understanding of AI capability.
The first issue I think is around recency and tier of use. I think a lot of people tried the free tier of ChatGPT somewhere last year and allowed it to inform their views on AI a little too much. This is a group of reactions laughing at various quirks of the models, hallucinations, etc. Yes I also saw the viral videos of OpenAI's Advanced Voice mode fumbling simple queries like "should I drive or walk to the carwash". The thing is that these free and old/deprecated models don't reflect the capability in the latest round of state of the art agentic models of this year, especially OpenAI Codex and Claude Code.
But that brings me to the second issue. Even if people paid $200/month to use the state of the art models, a lot of the capabilities are relatively "peaky" in highly technical areas. Typical queries around search, writing, advice, etc. are *not* the domain that has made the most noticeable and dramatic strides in capability. Partly, this is due to the technical details of reinforcement learning and its use of verifiable rewards. But partly, it's also because these use cases are not sufficiently prioritized by the companies in their hillclimbing because they don't lead to as much $$$ value. The goldmines are elsewhere, and the focus comes along.
So that brings me to the second group of people, who *both* 1) pay for and use the state of the art frontier agentic models (OpenAI Codex / Claude Code) and 2) do so professionally in technical domains like programming, math and research. This group of people is subject to the highest amount of "AI Psychosis" because the recent improvements in these domains as of this year have been nothing short of staggering. When you hand a computer terminal to one of these models, you can now watch them melt programming problems that you'd normally expect to take days/weeks of work. It's this second group of people that assigns a much greater gravity to the capabilities, their slope, and various cyber-related repercussions.
TLDR the people in these two groups are speaking past each other. It really is simultaneously the case that OpenAI's free and I think slightly orphaned (?) "Advanced Voice Mode" will fumble the dumbest questions in your Instagram's reels and *at the same time*, OpenAI's highest-tier and paid Codex model will go off for 1 hour to coherently restructure an entire code base, or find and exploit vulnerabilities in computer systems. This part really works and has made dramatic strides because 2 properties: 1) these domains offer explicit reward functions that are verifiable meaning they are easily amenable to reinforcement learning training (e.g. unit tests passed yes or no, in contrast to writing, which is much harder to explicitly judge), but also 2) they are a lot more valuable in b2b settings, meaning that the biggest fraction of the team is focused on improving them. So here we are.