Ben Papillon

“Everyone can code now!” Dude, no one can code now.

@tenobrus nature is healing

being a founder is reaching status lows i have never before seen my entire time in tech

Congrats to portfolio company @GetSchematic on their $6.5M raise, which we were honored to be a part of 🥳 Schematic builds entitlements and enforcement infrastructure for SaaS and AI companies. Put more simply, it serves as a digital gatekeeper for software and AI companies.

a little underwhelmed by opus 4.7 so far. I don't think it's bad, but in ~a week of heavy usage I haven't really noticed that it's better than 4.6, and it seems to make more erroneous left-field suggestions in planning that stem from context mistakes

@trq212 I'm going to try this, the big context windows scare me

You can also set your autocompact threshold yourself and effectively lower your context window if you'd prefer. For example, 400k context is a good compromise: CLAUDE_CODE_AUTO_COMPACT_WINDOW=400000 claude see docs here: #environment-variables" target="_blank" rel="nofollow noopener">code.claude.com/docs/en/settin…

I edited the intro because I realized I buried the lede originally- The 1M context window is a double-edged sword. It allows Claude to do more complex tasks but it can also leads to more context pollution if you don't manage your session well. This is how you do that:

he acts exactly like claude opus 4.6 high effort

@tekbog markdown

has there been a db winner for the AI era? next supabase or mongodb

@peterwildeford meta is that close you think?

Based on the data I see, I think: - Anthropic🇺🇸/Google🇺🇸/OpenAI🇺🇸 all ~tied - Meta🇺🇸 / xAI🇺🇸 each ~7mo behind - Moonshot🇨🇳/- Deepseek🇨🇳 / zAI 🇨🇳 / Alibaba🇨🇳each ~9mo behind - Mistral🇫🇷 ~1.5 years behind - No other companies competitive

@_colemurray @nateberkopec SOC-2 has entered the chat

for those balking at the idea of auto merging a PR, you are overestimating how much attention your employees/teammates are giving code reviews rely on systems (CI/monitoring), not someone’s LGTM!

skills, hooks etc are great but I'm amazed how much mileage you can get out of Claude Code simply via strategically-placed markdown files

@fleetingbits infra but perhaps isn't "softwere" per se communication software (VC, chat, etc) maybe a good example - I'd expect disruption to have more to do with user expectations changing (your points 6-7) vs changes to the value of software expertise (your earlier points)

@ben_papillon can you say more? what kinds of software do you consider to be more complex in the way that will make them more defensible?

some thoughts on saas companies and coding agents 1) i think that the threat of claude code and codex to a lot of saas companies is that it commoditizes their software expertise 2) most saas companies are more software specific than they are vertical specific; much of their expertise is in things like using cloud services and building data pipelines; sometimes they have some niche technical work 3) these used to be medium-hard problems to solve at scale; not so hard that other companies couldn’t do them, but hard enough that it was not worth trying to rebuild it all for a domain that already had a vertical saas company 4) the problem is that coding agents like claude code and codex commoditize these medium hard problems, which were the domain of saas companies; a small startup can now build all of this for a domain in a short period of time 5) on its own, this might be fine; it’s hard to compete with an existing company that already has sales traction in a domain; saas companies have customers, often with committed subscriptions, and can now afford to dramatically cut their R&D costs 6) but, at the same time there are real differences in what customers want out of products today, older saas companies were more about organizing data and creating managed workflows that let people structure their work 7) ai products instead replace the work and create the final work product and then give the user tools to verify the output; this means that there is a real difference now in what constitutes a good product, and this is an opportunity for new market entrants 8) importantly, doing the later is something that requires a different core competences, you need experience thinking about ai and ai interaction patterns, vertical domain experts on staff in professional domains, and sometimes ai enablement 9) the vertical domain experts are necessary to evaluate the ai work product and verification tools and the enablement is necessary to help customers fit ai into their existing workflows 10) this means that existing saas companies are in an interesting place; their existing revenue is probably solid on the short term and their ebita can be dramatically improved but their core competence has been commoditized and they have the wrong staff for the next phase of businesses 11) they suddenly feel more like pe plays than like vc plays and it’s hard to imagine that those saas companies with 30% year over year revenue growth will be able to maintain it 12) agents also have the possibility to reduce switching costs by sitting between users and individual products, reducing the human barrier to switching, and by making it easier to port your data from one product to another, even if vendors want to resist 13) and, i think there is another issue for companies that now want to sell software infrastructure products consumed by developers; it is easier for your customers to build a version of it themselves 14) and, if claude code just builds a substitute of your product for the user, the customer may never look up your service to purchase it, even if it is marginally better, the build to buy calculus has been now made more complicated by build becoming lower friction 15) i think this means that there will need to be developments in making purchasing easier and discoverability of new software products easier; this probably needs to occur through llm chat services; and, it will probably be a great revenue stream for openai and anthropic 16) note, this may also be an issue for saas products that don't use a lot of domain knowledge and which are consumed by startups, which are more inclined to build themselves, but this is less obvious

who is building `npx skills add ballmer-peak`

@garrytan ask claude code about the ballmer peak

I'm giving up drinking because of Claude Code. I need my brain to be maximally pristine so I can sling 10k LOC a day

@tenobrus glad to see him escaping the permanent underclass

Donald Knuth is vibemathing now. real tough day for the stochastic-parrot crew.

@thenanyu I’ll never forget Conan O’Brien’s pronunciation of “B2B SaaS” during such an ad read

Hearing Bill Simmons ad read for LinkedIn Recruiter and having absolutely no clue what any word means

another hill climbed

Imagine you could ask Claude anything about how your customers use your product, how much they pay, and how close they are to their limits. Ok, wish granted. We just shipped a Schematic MCP server for Claude Code. github.com/schematichq/sc…

@GergelyOrosz I would imagine their infra strain vastly outweighs any other factors

On one end, the Anthropic team is a massive user of AI to write code (80%+ of all code deployed is written by Claude Code). They ship amazingly fast. On the other hand, seeing these beyond terrible reliability numbers suggests there might be a downside to all this speed:

@dillon_mulroy @dexhorthy haven’t heard that name in years

subagents are sparkling map reduce