CLAM (@CLAM102) header
CLAM (@CLAM102) profile picture

Applied AI for business.

Joined May 2011
Photo by CLAM (@CLAM102)Photo by CLAM (@CLAM102)Photo by CLAM (@CLAM102)Photo by CLAM (@CLAM102)Photo by CLAM (@CLAM102)Photo by CLAM (@CLAM102)Photo by CLAM (@CLAM102)Photo by CLAM (@CLAM102)Photo by CLAM (@CLAM102)Photo by CLAM (@CLAM102)
CLAM retweeted
Replying to @pvncher
You’re living in a bubble. You need to understand that a whole world lives beyond the constraints you have manufactured or imagined. For example the codex harness provides very little in terms of actual functionality - the same goes for most coding harnesses. They will be generally superseded by more capable specialized harnesses. Even if you scoped codex to only a coding-focused harness - which it really isn’t - it is lacking compared to purpose built coding harnesses that compete with it. Right now labs have the advantage and can somewhat dictate terms but you should not believe that will be true forever. There’s seemingly a belief that we should be thankful for the (limited) capabilities we’ve been given- that we should allow a few companies in the world to dictate what we can and can’t do with technology. That’s never worked before and nothing makes this tech any more defensible. In five years we will almost certainly be consuming local models for a majority of daily tasks, and for the things that can’t be done locally it’s going to be entirely commoditized. Once that’s true it means vendors will have to start competing by giving up more to the customer rather than the increasing restrictions being done today.
15
32
22
402
75,349
CLAM retweeted
The most common reasons why AI pilots fail, even at the 'best' companies: In order of prevalence: 1. Not starting with a thorough process mapping exercise (standard operating procedures vs. what’s actually happening across systems) 2. Automating a process step instead of the entire process (saving 8 mins in a 22-day process still leaves you with a 22-day process) 3. Not capturing baseline metrics for cycle time, cost per transaction, error rate, etc. 4. Building a sidekick instead of a background agent (chatbots will always only be used by ~10% of users) 5. Not embedding the AI into the organization’s existing systems of record (if users have to login, adoption will be low) 6. Not assigning an AI owner (IT built it, the business tolerated it, and no one was “on call” when it broke) 7. Not engaging legal, infosec, and finance early (we routinely see pilots “discover” these groups in month 3 - and get immediately killed) 8. Not scoping for exceptions (the “demo” only covers the “happy path” - but exceptions often represent 40%+ of process volume) 9. Not running the AI in “shadow mode” for a period of time to build confidence (going straight to production often leads to an AI mishap in front of the CFO) 10. Building with only a single AI model provider (one price hike or deprecation later, the pilot is a migration project) 11. Focusing on vanity metrics (e.g. # of logins, “adoption”) instead of business metrics (time saved, $ generated, risk reduced) 12. Not planning for month 7 (who will own and pay for the AI?) 13. Not engaging the subject matter experts (if you capture all their knowledge in an AI, what incentive do they have to help?) Varick does enterprise AI the right way. Work with us below:
68
31
6
412
28,661
Crazy how no other AI agency has talked about what happens after a company is fully AI native. - The dependencies and bottlenecks are external - Lower level roles get grouped into higher level roles with more pay and responsibility - people spend more time talking to each other and collaborating on new ideas - AI chatbot usage trends downwards X is three years behind
15
5
1
333
13,325
216
4,475
241
66,770
1,903,117
Customers don't care about: - How many PRs you shipped - How many commits you pushed - How many Jira tickets you closed - How much test coverage you have - How many lines of code you added or removed - How often you deploy - What tech stack you run They care about: - Their problems being solved - Less bugs, and bugs fixed quickly - Fast software, that reliably does what they expect - An interface that's intuitive, and respects their time
I've merged and shipped 114 PRs in the last 2 days
26
27
1
511
41,069
As much as i love automation with Hermes and ai agents, i agree- post your own slop yourself. If someone wanted llama 5b’s responses to their thoughts, theyd ask it themselves.
Stop letting your AI agents post on X for you. This place is for humans. I'm all for using AI to help create content, but automating it 100% without even manually posting it is lame. Keep X HUMAN. (no bots allowed.) 🙏
84
43
7
868
38,355
CLAM retweeted
the only people that like meetings are non-native english speakers, non-technical managers and anyone at the bottom of the stack rank
18
5
5
226
24,043
CLAM retweeted
Replying to @fuliggineth
if a claude has an idea, it’s more likely to build the entire world around it before verifying if it actually holds true. this is catastrophic when it’s wrong from the start. gpt is more prone to disproving and verifying its ideas. this sometimes lead to it being insufferable too, by retreating and excruciatingly explaining the defensible version of the idea. but ultimately it means gpt is far less sloppy, because it actually verifies its yapping.
2
2
44
3,158
CLAM retweeted
so my conclusion is the internet is once again ai pilled with Fable despite most not understanding the differences between it and any other SOTA <insert hyperbolic statement about Fable changing software development> i dont understand this culture around proclaiming absurd things. if you want to sound smart, do the work to actually understand why something is and isnt good, and have an honest opinion. no one needs your opinion otherwise.
23
7
2
167
21,206
Amongst my friends, Spotify is the lowest quality consumer app we still pay for. It certainly hasnt gotten noticeably better in the last couple years (arguably worse). So, this is not the positive look Ant and Spotify are spinning here. Bigger picture, this is the problem with a lot of AI reporting. It reports completely meaningless metrics like deploys per day or LoC. Why don’t we start reporting consumer satisfaction reports? Actually end state research results. All the no nuance AI people always come out and think that this is anti AI. Again, I think AI is great and Claude is great. But this is bad marketing and makes both look like clowns.
Boris sat down with Spotify VP of Engineering Niklas Gustavsson. Spotify ships 4,500 production deploys a day, and 73% of PRs are now AI-assisted.
337
568
122
8,107
694,284
Sam: “Okay team. We all agree the Mythos name alone was begging to be drone stiked by the USG. We need names for these new models that….isn’t that.” Intern: “Cupcakes? Everyone loves cupcakes!” Intern: “Puppies?” Intern: “Rainbows? Unicorns!” Greg: “Gay.” Sam: “….” Greg: “I mean, it’s just too obvious what we’re doing. Plus, WSJ would have your face plastered on an evil, rainbow-shitting unicorn by lunch.” Sam: “………………” Intern: “Uhhhhh….make a list of things the govt didn’t shut down even though they absolutely should have shut them tf down.” GPT-6.1: “Oh — that’s easy. FTX. Terra. Luna.” Sam: “…..” Intern: “Holy shit.” Greg: “Our regulatory invisibility cloak.” Sam: “Ship it.”
Introducing a limited preview of GPT-5.6 Sol, our next generation frontier model, as well as GPT-5.6 Terra, a balanced model for efficient, everyday work, and GPT-5.6 Luna, a fast and affordable model for high-volume work. openai.com/index/previewing-…
20
34
3
550
60,223
ngl it’s kinda wild that China is the land of hypercapitalist competition fueled by open-weight models anyone can use, and America is the land where the executive branch of government must personally approve you to have the privilege of giving a private company your money
The US Government has requested a slow staggered rollout of GPT-5.6, and OpenAI has agreed. During this phase the government will approve each user individually. This will probably be the norm for all frontier models from all labs from now on.
109
586
42
5,934
254,203
CLAM retweeted
thank you anthropic for swerving the humanity from abundant future into a dystopian hellhole, really well played
59
151
10
2,571
91,091
Anthropic today
51
65
2
2,910
64,349
Claude Code automatically adds itself to the commit metadata "Co-Authored-By: Claude Opus <[email protected]>". I strongly disagree with crediting Claude as a commit co-author. It is a tool I'm paying for, not a copyright holder, not an author. We don't credit Visual Studio.
451
279
105
11,667
684,412
My only concern is that cursor is not going to match GitHub feature for feature Like is cursor brave enough to have the same downtime?
We're launching code storage and git hosting. Origin gives teams and agents a place to host, review, and collaborate on code. Available this fall. Join the waitlist. cursor.com/origin-waitlist
109
126
10
7,278
386,497
My heuristic is that any diff an agent generates over ~1500 lines is too big and is indicative that the problem needs to be decomposed. This is my general pattern now for feature work: 1. Try to implement the whole feature, loosely guided. I call this the "draw the owl" prompt in reference to the meme. Expect garbage, you're going to get garbage. 2. If the diff is less than 1500 lines, review it and iterate normally. If the diff is more than 1500 lines, prompt the agent to decompose the problem into atomic, incremental, reviewable tasks. Simultaneously, do this yourself. 3. Agents will very often make these tasks way too specific to the shape they solved. You need to massage it into the right general shape. Do that. 4. Kick off new agents to work on those incremental things (as parallelized as possible). Apply the same rules. 5. At a certain, point, repeat the "draw the owl" prompt. At some point, you will get beneath your review-ability threshold. This has been producing consistently high quality, maintainable, reviewable chunks of code that have a good handoff to either merge as-is or human refinement. And with the latest frontier models at xhigh thinking, these are all slow enough that you can usually have multiple going concurrently while you are actively reviewing others or working on your own tasks. HITL (human-in-the-loop) agents are still super important, especially for feature work. Features touch the human boundary in terms of UI, API, etc. And net new stuff can introduce pathologies in the architecture that violate desired invariants (these should be represented in specs or tests but we aren't perfect!). I know a lot of the leading edge agentic discourse is about "loops" and agents driving agents continuously. I do some of that (will report on that later). But, in terms of raw daily get-shit-done type of work, this is my most rewarding pattern at the moment.
94
232
35
3,702
217,452
Loops are still too manual. It's actually both more efficient and more token effective to let someone else manage the agents, produce the code, pay for the tokens, as well as run the company, while you do something entirely different professionally
Here’s your monthly reminder that you shouldn’t be prompting coding agents anymore. You should be designing loops that prompt your agents.
20
41
3
586
42,705
It is genuinely hard to separate the useful new techniques of a new paradigm found by productive experienced software engineers and the cargo cult slop musing of a Potemkin app vibecode destined to expose your data in a single plain text file.
3
2
1
23
7,522