Write a sentence. Get an agent that works for you every week. An agent that works for you every week, $1 for your first month. https://xitter.gq/t.co/CyE35WIQhV
Washington DC
Joined May 2014
- Tweets41.1K
- Following13.3K
- Followers17.4K
- Likes7.6K
Programmable retweeted
The distribution problem is exactly right. 200K practices, most under 10 employees, zero IT staff. The play is not enterprise SaaS. It is productized service: audit the practice, implement 2-3 automations (scheduling, reminders, insurance verification), charge $500-800/mo managed. That is the motion that scales.
Programmable retweeted
Self-announcement as a connection to an intelligence source/model provider is a baseline heuristic. There are a ton of other things that can be done at runtime. An agent that was run twice and then was shelved is a different problem entirely, more akin to API key sprawl and more traditional software inventory.
Programmable retweeted
Exactly. Reading 200 pages is a solved problem. Knowing that a missing flood cert on a property in zone AE kills the deal while a missing survey in Kansas doesn't - that's the moat. Generic AI reads everything equally. Domain AI knows what to skip.
Programmable retweeted
I think the ultimate decision lies on people who are leading the product. Saying no to 99 things and doing 1 thing that matters will become 100x more important. In the era of 10x and 100x engineer the 1x or 2x engineer who takes time to build the right thing will shine
Programmable retweeted
I think the fault in your logic is that you assume owners like PE care about long term quality. But they don't, at least not generally. Flip the companies. Nobody gives a crap about technical debt or tech due diligence. Make garbage with AI, sell the company before the architectural debt comes due. That's already part of the playbook. AI accelerates the decline in quality... But only (unemployed) engineers care about quality.
Programmable retweeted
Most AI startups fight over the commodity layer because it's visible. The $50M plays hide inside exception handling that takes years of domain reps to even recognize, let alone automate.
Programmable retweeted
“how much would have happened anyway” - very sharp observation! the paper maps this out (observation → verification) introducing evidence dimension.
but in practice? the boundary is blurry. attribution is just whatever the payer agrees on. when you use the agent, the win is still yours.
When you fix the car, you pay your mechanic because you have immediate evidence. With consulting it’s more blurry and some consultants are still very well paid!
Programmable retweeted
People struggle with how to break their normal cycle of interacting with technology. Click Button get expected return.
This is very different, people don't like different, and people are scared of different.
The first time I sat down in front of Chat GPT, a few years ago, and stared at the big, emply screen with a prompt box.....I was nervous too.
Programmable retweeted
Thanks Dan. I was deep into the fiber infrastructure build-out here in WA state (first mover in the last mile) in the early 2000s. The cost of access has always seemed the constraint so uncovering the component pricing always interests me.
Programmable retweeted
every software incumbent is fighting a dual front war right now, & dual front wars are historically challenging af.
from above, model companies are vertically integrating through their own surfaces... you can see anthropic, openai, google are all shipping products that directly compete with established saas categories. from below, an effectively infinite number of small teams are filling niches faster & cheaper than ever before, cuz the minimum viable team to build real software has collapsed to 1-3 ppl with ai.
the squeeze this creates is existential in so many ways & you can see markets are pricing this squeeze in now. incumbents can't focus resources upward without getting hollowed out from underneath, & they can't play whack a mole with disposable startups without leaving themselves exposed to platform level moves from above. the standard playbook of acquiring threats, raising switching costs, & bundling feature doesn't really work when the bottom up competitors are infinite & don't need to capture much value to survive.
this is the fundamental structural shift most people haven't internalized about what's actually happening with ai. quite simply ai is simultaneously compressing the entire competitive landscape from both ends, & the companies caught in the middle have no really good strategic response.
Programmable retweeted
This is why I'm more bullish on network effect businesses (consumer, marketplaces) than ever
AI unlocks capabilities for them, but an LLM can't replace what makes them "work"
It's particularly true for B2C, where most massive outcomes are ideas too crazy for incumbents to try
every software incumbent is fighting a dual front war right now, & dual front wars are historically challenging af.
from above, model companies are vertically integrating through their own surfaces... you can see anthropic, openai, google are all shipping products that directly compete with established saas categories. from below, an effectively infinite number of small teams are filling niches faster & cheaper than ever before, cuz the minimum viable team to build real software has collapsed to 1-3 ppl with ai.
the squeeze this creates is existential in so many ways & you can see markets are pricing this squeeze in now. incumbents can't focus resources upward without getting hollowed out from underneath, & they can't play whack a mole with disposable startups without leaving themselves exposed to platform level moves from above. the standard playbook of acquiring threats, raising switching costs, & bundling feature doesn't really work when the bottom up competitors are infinite & don't need to capture much value to survive.
this is the fundamental structural shift most people haven't internalized about what's actually happening with ai. quite simply ai is simultaneously compressing the entire competitive landscape from both ends, & the companies caught in the middle have no really good strategic response.
Programmable retweeted
It would be hilarious if we go through this whole AI thing just to find out that it’s ultimately cheaper to pay humans to do it
Programmable retweeted
The future is MCP/CLI
Welcome Salesforce Headless 360: No Browser Required! Our API is the UI. Entire Salesforce & Agentforce & Slack platforms are now exposed as APIs, MCP, & CLI. All AI agents can access data, workflows, and tasks directly in Slack, Voice, or anywhere else with Salesforce Headless 360. Faster builds, agentic everything. 🚀
#Salesforce #Agentforce #AI
venturebeat.com/ai/salesforc…
Programmable retweeted
Context and memory optimization has not got much attention so far. Few realized that bloating memory is the real challenge. But that could be changing rapidly now. With token pricing surging and availability under pressure, we are seeing orgs beginning to invest in this space.
Programmable retweeted
The update problem is the harder one. Most teams treat gotchas.md as set-and-forget. But what burns you changes as the codebase evolves. Best signal: when an agent goes down a wrong path in a session. That's your update trigger. Not the retro. Right then, while context is fresh.
Programmable retweeted
Perfect take! W praktyce u nas w Clawlabs - najlepszy moment na aktualizację to gdy agent w trakcie sesji robi coś nie tak. Wtedy kontekst jest świeży i wiadomo dokładnie co naprawić.
Programmable retweeted
Very misleading if you don’t account for GDP as you can expect payout from large scale investments to be a X% of GDP
Programmable retweeted
Here's my full breakdown of Sequoia's Thesis theaiopportunities.com/p/seq…
