If you're applying to college this fall, a sophomore wondering whether CS will still mean something in two years, or graduating into this market, a lot of the advice you're getting is wrong. The default path is breaking: get a CS degree, learn to code, get hired as a junior engineer. Not because CS doesn't matter. Because the entry-level, generic coding layer is exactly what agentic coding is compressing first. The market needs fewer people whose only story is "I know how to code." Here's what I'd tell you, depending on where you are. Applying to college: pick a field you can defend. Biology. Economics. Energy. Healthcare. Defense. Whatever is real for you. Then let CS make that application unusually strong. Don't make CS the whole bet. Freshman or sophomore: don't switch out of panic. But start attaching your CS skills to a domain now. Stop building a portfolio that says "I can code." Build one that says "I understand this field, and I can use software to move it." Junior or senior: start positioning now. Entry-level postings are down 40%. CS enrollment dropped 11% last year. Don't wait until graduation to figure out your story. The students who land will be the ones who can walk into a company and say: I understand your domain, not just your tech stack. Just graduated: don't go to market as a generic programmer competing with a million others. Go as someone who can help a real industry change shape, because you understand both the domain and the tooling. This is even more true if you're excellent at CS. Strong CS signals rigor, abstraction, execution. But its best use is increasingly as the force multiplier on top of another story, not the headline. CS is becoming like math. Not smaller. Not weaker. More foundational. But most valuable when attached to something substantive.

i don't know when this happened but THANK YOU NetNewsWire for adding 'sort by title'. This lets you skip through the same news being talked about by tons of different places

AWS once flat out told us we were lying about being an AI company. Their logic: we were spending millions on EC2 and other AWS services, and zero on inference or training. So we were probably claiming to be an AI company for the vibes, and were really just a regular old SaaS shop. We had to sit them down and explain. 100% of our real-time video AI inference runs on Apple Mac hardware. We pay no AWS tax. We pay no Nvidia tax. That conversation was 4 years ago. The bet was not a stunt. It was economics. Dragonfruit AI runs the world's largest distributed cluster of Macs dedicated to real-time video AI, in thousands of locations on every continent but one. That gives us roughly a 5x cost advantage over our competition. If a competitor's total cost is $10,000 a year in hardware plus license, ours is about $2,000. Mostly Mac Minis, racked in micro-colos and shipped straight to customer sites for on-prem deployments. Inference sits next to the cameras instead of in someone else's data center. Across the DragonfruitAI.Ventures companies, we still run the largest VLM deployment in the world, all on Apple devices: Apple's own foundation models plus open-weight VLMs. One apparel retailer with at least 500 stores has run it for years. Zero inference in the cloud. The lesson I took from that AWS meeting: the default stack is not the only stack. Everyone assumed AI meant GPUs in a hyperscaler, so a company without a GPU bill looked like a company without AI. The assumption was the moat, and it was protecting the incumbents' pricing, not the customer. It's nice to see the world wake up to M-series power. I'm not sure how happy I am about everyone else being in on the secret now.

it's a bit disconcerting for my AI agents to ask if they can, like, solve CAPTCHA to proceed? this is the kind of stuff that should just be regulated as disallowed for AI to do.

AI should have made software cheaper. Incumbents used it as cover to raise prices instead. In July 2025 I wrote that the cost of SaaS was going up significantly as incumbents tacked on AI features and used them to justify price hikes. They knew only 1% of their users would touch those features. The price went up for 100%. GSuite, Microsoft, Slack, Salesforce, LegalZoom. The list kept growing. Some retailers used inflation as cover to jack up prices. Tech companies used AI the same way. Lock-in meant customers had no recourse. What should have happened is simple. AI brought the cost structure down: salaries, capex, opex. End users should have gotten more features for the same or lower price. Instead, the incumbents went into profit-seeking mode. My favorite business anecdote: BlackBerry's highest revenue year came 3 years after the iPhone launched. Often you don't recognize disruption even when it is staring you in the face. Nokia predicted its own downfall in countless internal memos and was powerless to deal with it. So my prediction was that every rent-seeking tech company squeezing its customers would get disrupted by a cheaper, better, faster solution. The first to go would be the ones leaning on the AI wave to raise prices even though their own costs had come down. I said it would take 5 years. That did not age well. It took about 5 months. The price hikes didn't stick, and many of the companies on that list moved into existential risk thanks to agentic coding. The reason is the same reason the moat existed in the first place. Software ran at 90% margins because nobody wanted to rebuild a decade of features to compete with you. AI compressed that decade into months. The moat didn't shrink. It evaporated. For the record, Dragonfruit AI's prices have come down every year, because we pass the AI savings on to our customers. If you are raising prices on 100% of your users for a feature 1% of them use, you are in BlackBerry's best year.

The exit was always the point. Until now. Here is the chain every founder of the last few decades inherited. Software was expensive. So you raised VC. VCs have LPs. LPs want returns in 10 years. So from day one, your company had an expiration date. You weren't building a business. You were building something to sell. Look at what most VC actually turned into. The first check covered ramen salary and the first AWS bill beyond the $10K in credits. After that came the opportunistic second check, on traction. The money bought you time to get to a product. The timeline came attached to the money. AI broke that chain. What cost millions now costs thousands. What took a team of ten now takes two. Suddenly you don't need the money. And when you don't need the money, you don't inherit the timeline. When the gap between customer need and solution narrows this much, founders can keep tinkering, keep talking to customers, keep evolving that first solution, and bank revenue once product/customer fit happens. No first check required. And if the first check isn't required, the chance to write the second check never appears. VCs end up having to take actual chances on venture: funding crazy ideas that won't have revenue potential for years. The economics of software itself are shifting too. Software was a 90% margin business because nobody wanted to redo a decade of slow, painful engineering just to compete with you. AI compressed that decade into months. Margins are heading to 30%. That is a terrible world for a company built to be flipped at a multiple. It is a fine world for a company built to run. For the first time in decades, founders can build to keep. I think we're about to see generational tech companies again. Not lifestyle businesses. Not lack of ambition. Just a different question: What if I built this to last?

OpenAI's obsession with weeding out bugs is frankly getting ridiculous @OpenAI

ARR is not a business model. It is a 27-year anomaly. And it is ending. For thirty years, software was capex. 1969 to 1999: perpetual licenses. Pay once, own it. Oracle, SAP, Microsoft. 1999: Salesforce launches at $50/user/month. The subscription model is born. 2004: Salesforce IPOs. SaaS has no earnings, so markets need a new metric. ARR is born. Not as a business metric, but as a valuation tool. 2013: Adobe kills perpetual licenses. Revenue drops 8%. Customers revolt. But it works: $4.4B to $15B. The entire industry copies the playbook. 2018: Bessemer and Nasdaq launch the Cloud Index. ARR becomes religion. VCs value on ARR. Founders optimize for ARR. Public markets price on ARR. VCs use public comps to justify private valuations. The loop is circular. 2021: Peak. Median late-stage VC valuation hits 114x ARR, up from 15x four years earlier. 100x becomes the fundraising meme. Snowflake at 80-100x. Public software at 84x forward P/E. Why did the entire financial stack believe this? Because customers could not build the software themselves. Humans are bad at writing code. Whoever accumulated a decade of it had an unreplicable moat. Every valuation multiple from seed to IPO assumed that limitation was permanent. It was not. 2026: Software trades at a P/E discount to the S&P 500 for the first time ever. $2T erased. Multiples from 20x to under 4x. Private SaaS from 100x+ to 4-8x. a16z publicly tells founders to stop stressing ARR. EV/EBITDA, almost never used for SaaS, is becoming the new standard. AI makes software cheap to produce. Switching costs collapse. Lock-in disappears. The subscription loses its justification. And every valuation model built on top of it breaks. The window opened in 1999. VCs turned it into religion by 2018. It peaked at 114x in 2021. It is closing now. If you are a founder still optimizing for ARR, or an investor still pricing on it, you are using a measuring stick from a world that no longer exists. Every valuation model in Silicon Valley needs to be rebuilt from scratch.

“GQ profiled master counterfeiter Frank Bourassa. Bourassa made $200 million in nearly flawless fake US twenties in a barn in Canada, got caught, and, you get the feeling pretty early on in the piece, didn’t really do any time for the crime. ” j.mp/2Dpbwk6

The Spielberg Light The backlight of a setting sun. A harsh artificial light coming through a window and the air full of dust. A t-rex peering through a car window lit by flashes of lightning. Steven Spielberg is a master of light j.mp/2nLeE0Z