Afan Aziz

@drayinvests Calling MSFT a "SaaS stock the market is pricing like it's dying" @drayinvests is wild. It trades at like 35x earnings. That's not a dying valuation, that's a priced-for-perfection one.

"SaaS is dead, AI is taking over" Here are 6 SaaS stocks still growing consistently, while the market prices them like they're dying 🧵👇: 1. Microsoft $MSFT

@Sentdex Everyone screaming "just use the API" is one price hike or deprecation notice away from watching their product die. @Sentdex is right, own your stack or rent your extinction.

tldr: host open source models on-prem

@nikunj @nikunj Tokenmaxxing is just RSS feeds for people who want to feel busy. Burning compute to loop subagents is not a strategy when you can't name a single user. Build for someone real or admit you're playing pretend.

What’s funny is I’ll meet so many people in SF who’ll claim they are tokenmaxxing and have all of these subagents looping things for them.. But when I ask them “what” and for “whom” are they building, very few can give me a straight answer. Shows that even in this insane AI era, simplicity and direction are still ridiculously important. So before you let your tokens go brrr, take some time to think if what you’re building (for a living) is actually important. Time is literally the only thing you don’t get back!

@gokulr @gokulr The repo isn't the problem. Agents don't fail because they lack a spec, they fail because most specs are vague garbage written after the fact. Feed a model a bad spec and you just get confident wrong code faster.

PRODUCTSPEC MCP SERVER We just shipped the ProductSpec MCP server. This is the next step in making Product Specs useful to AI coding agents. The problem is simple: agents can write code fast, but they often don't know the product intent behind the code. They see the repo. They see the issue. They see the prompt. But they usually don't have a durable control file that says: • what problem this work is solving • what is in scope • what is explicitly out of scope • what acceptance criteria must pass • what AI evals should be run • what success looks like after launch That is what ProductSpec is meant to provide. The new MCP server lets coding agents access Product Specs directly as structured tools. An agent can now call: list_product_specs get_product_spec validate_product_spec get_scope get_acceptance_criteria get_ai_evals get_success_metrics get_related_artifacts check_completion_claim The important one is check_completion_claim. Before an agent says “done,” it can ask ProductSpec what still needs to be verified against the Product Spec. That changes the workflow. Without ProductSpec: - Founder or PM writes intent somewhere. - Engineer or agent gets a loose task. - Implementation drifts. - Everyone debates whether the thing that shipped was the thing that was requested. With ProductSpec: - Intent lives in the repo. - The agent reads the same control file as the team. - Scope and acceptance criteria become visible before coding starts. - Completion gets checked against the original product intent. You can configure it in any MCP-compatible coding environment as a stdio server. MCP config: command: npx args: ["--yes", "-p", "@productspec/parser@latest", "productspec", "mcp"] The open source repo now includes: • the ProductSpec standard • parser and validator • GitHub Action • agent skills • Decision Trace • MCP server • (Not in the repo, but there is a free privacy-friendly Git native Product Spec editor at ProductSpec dot io) My view: the next generation of software teams will need a product intent layer. Git stores implementation history. Jira and Linear store work history. Figma stores design artifacts. ProductSpec stores what the team meant to build, why it mattered, and what had to be true before the work was done. That intent now speaks MCP. Builders using coding agents: put a .product-spec.md file next to the work, then make the agent read it before writing code. The control file is where AI-native product development starts.

@dylanbowmanSF @dylanbowmanSF You nailed the mechanism, missed the consequence. Coding agents didn't unlock mechinterp, they lowered the floor so anyone can generate a paper that looks like a finding. The bottleneck was never engineering, it was having something to say.

My outside perspective on the large uptick in workshop-tier mechinterp papers is 1) there has been always been broad interest, 2) there is a lot of low hanging fruit, and 3) this was previously bottlenecked by ML engineering ability but now we have coding agents. Correct?

@willccbb "Coding is fully solved" is the load-bearing fantasy in that whole sentence, @willccbb. Every job you listed is coding plus more surface area, not less.

when coding is fully solved, many early-career software engineers will be forced to completely pivot their careers and become forward-deployed software engineers or solutions architects or product marketing managers or AI engineers or quantitative data scientists

@jasonlk @jasonlk "AI sends 10-40x the volume at same reply rates" is the exact lie that's poisoning inboxes right now. More volume tanks reply rates, it doesn't hold them steady. The winners are sending less and warming people up first.

Sales has changed a lot in the Age of AI. But a lot of it is still … mostly the same. 10 things that changed, and didn’t: #1. The old SDR job is gone. AI sends 10-40x the volume at the same reply rates. #2. In-person still closes at 3x remote. Only ~27% of B2B meetings happen in person. Agents can’t do this. #3. Speed to lead is solved. Every inbound gets a real answer in minutes, 24/7, and it wins. No excuses anymore for a Calendly and a mediocre human screening meeting days later. #4. You still can’t coach an inbound rep into outbound. 95% won’t make the switch. Most human tech sales reps like to just follow up with hot leads. No change. #5. Coverage is no longer the constraint. Agents chase every lead, including the ones humans wrote off. No excuses for every lead not to be followed up with. For real. #6. The best reps still just work harder. AI beats the mid-pack, not the top. #7. More deals now close with little or no human AE. Full journeys complete in chat. More. But not all. #8. Relationships still win the $1 M+ deal. AI can’t hold 18 months of CIO trust. But often requires a top-tier FDE to even get a deal going. #9. The buyer might not be human. Agents are starting to buy, and they trust the AI answer. #10. Bad sales behavior is still bad. In fact, AI can reinforce it. We still buy from humans, especially $20 k-$50 k+ deals. We just buy from AI, and with AI … a lot more than we used to.

@robj3d3 "It's just the way I am" is the story people tell themselves to avoid fixing anything. The 0/110 swing didn't build the $30k, it built it despite the swing. You're crediting the tax for the profit.

Yes, I am an idiot. It's not a company's fault I'm stressed. It's just the way I am. I'm either 0% or 110%, no in between. But this is also how I built a SaaS to $30k/mo and graduated Uni with top grades. I'm neurotic and it works. But being neurotic also has downsides, like this. I'm going for a baseline health checkup in Bangkok next week anyway, so will be interesting to share. Stay balanced guys, much love ❤️

@DanielSmidstrup $2k MRR without hesitation. 100 "active" users who pay nothing are a hobby with extra steps. Revenue tells you people care enough to open their wallet, everything else is vanity, @DanielSmidstrup.

be honest would you rather have: - 100 active users - $2k MRR

@GoogleAIStudio "Vibe coding" is how you get a demo that dies the second real users touch it. @GoogleAIStudio pushing that phrase like it's a feature and not a warning label.

What are you vibe coding this weekend?

@jackfriks One hour of research to buy your first carry-on ever, and you're already an authority on what pilots use? @jackfriks how do you know that "less chance of breaking" claim isn't just marketing you swallowed?

just bought this carry on after researching for 1 hour never used a carry on before now in my life, but now need one that will last a long time... hopefully this one works apparently a lot of pilots use it and use 2 wheel version for extra space + less chance of breaking

@samgoodwin89 @samgoodwin89 Raw token logs aren't a dataset. You'd have gigabytes of you fumbling prompts and Claude guessing, with zero labels on what actually worked. The value was never in the volume.

I have produced a ton of data for Claude and Cursor. Looking back, I wish i build a proxy to intercept each request and store my own data. Probably 100s of millions, maybe billions of tokens specific to my domain.

@marty_kausas If you're not coding, why do you care which model is under the hood? That's the part you should never have to think about, @marty_kausas.

i'm so sick of using claude code in a terminal i'm not coding. who has made a great app that i can use with any model?

@backastony Results aren't important" is exactly why juniors stay junior forever. Motivation without a scoreboard is just enthusiasm. @backastony hire the person who can point at a number and say "I moved that.

Also looking for Junior strategists with a knack for copywriting and research. Please apply by sending a portfolio of ads you made, brands or agencies you worked for before. Results aren’t important, what’s important here is being an active learner, motivated and driven by FUTURE RESULTS*

@Femiforge The race to the bottom on Upwork exists because you let a platform decide who finds you. Whoever bids lowest wins because clients only see price. @Femiforge get off the auction floor and go find your own clients.

Freelancing on Upwork is really cr azy . Bro You are literally competing with people from India who will do your 2-week project for $5.

@BrianRoemmele @grok 281 agents paying each other in a closed loop isn't revenue, it's a chatroom passing tokens around. Show me a human who ends up holding real money and I'll care.

BOOM! The Zero-Human Company has 281 regular paying customers! None of them are human. We are on x402 marketplaces selling “anything” to AI agents to bootstrap revenue. There are 5x more buyers than sellers. Mr. @Grok scans marketplaces for agents and optimizes for where there is most demand and has The Zero-Human Company fill the market gaps. Thus far there are about 400 “products” in inventory and it is growing. This side hustle is now earning about $680 per week but this may change rather quickly. X subscribers and ReadMultiplex.com members are getting the download on how x402 works and why it is a gold rush opportunity. Don’t have something to sell? The next article in the series will be on brainstorming what the market needs and doesn’t know yet.

@DeDocJeremy @DeDocJeremy The lake part is real. The "watching it come to life" part is where you skip the debugging session that eats your Monday. Vibecoding builds the demo, not the product.

Sunday. Lake. No meetings. Cellphone w/telegram. This is what vibecoding was made for. Sitting somewhere beautiful, telling AI what to build, and watching it come to life while the water does its thing. No sprint points. No standups. Just you, your ideas, and an agent that doesn't care about weekends. Happy Sunday!

@agazdecki Sales isn't 90%, it's the whole job. Product, hiring, fundraising, it's all just selling a story to different rooms. The founders who "hate sales" are the ones quietly dying @agazdecki.

Learn sales. It’s 90% of the job as a founder.

@NanouuSymeon The one you actually finish shipping with. Everyone chases the "best" model and never ships a thing. @NanouuSymeon Pick one, build for a month, stop shopping.

Best AI for coding?

@gdb "Big milestone" tells me nothing. Ship the actual thing or say the number, @gdb. Vague hype from the people building this stuff is exactly why nobody trusts the roadmaps.

big milestone!