Akashvani: 2025 | December
Philosophy, AI Bubbles, Agentic Shoppers, and Why God Doesn't Shave
Hi -
I hope you’re well.
I wanted to share a few things I’ve come across that might be helpful – Philosophy, AI Bubbles, Agentic Shoppers, and Why God Doesn’t Shave. Let’s get going.
1. Instead of starting with tech this time, let’s start with something that should be top of everyone’s list – how to live. As some of you know, I’ve been on this kick to read the classics. This book (”The Practicing Stoic“) just makes it easier – the author takes different categories (Judgement, Externals, Perspective, Death, Desire, Wealth and Pleasure, What Others Think, Valuation, Emotion, Adversity, Virtue, Learning -- pretty much what all humans go through) and then has commentary and quotes from the legends (Seneca, Cicero, Plutarch, Montaigne etc.) on those topics. Wonderfully compiled and great food for thought. There’s so much here for the 🧐thoughtful reader, but a few things that stood out for me.
· “We believe these affairs of ours are great because we are small” - Seneca
· “If we were all to bring our misfortunes into a common store, so that each person should receive an equal share in the distribution, the majority would be glad to take up their own and depart” - Plutarch
· “I will lead you by a shortcut to the greatest wealth…My dear Lucilius, not wanting something is just as good as having it” - Seneca
· “The man in possession of himself knocks in vain at poetry’s gates” - Plato
· “And in truth, he who depends on the caprice of the ignorant rabble cannot be numbered among the great” - Cicero. [Presidential, no?]
· “How foolish one must be to leave a lecture hall gratified by the applause of the ignorant! Why do you take pleasure in praise from those you cannot praise yourself?” - Seneca
· “Some people obtain fame, and others deserve it” - Lessing via Schopenhauer
· “What is it to be insulted? Stand by a stone and insult it; what will you gain? And if you listen like a stone, what will be gained by the one who insults you? - Epictetus. [Epictetus is one of my favorites!]
· “If you hear that someone has spoken ill of you, do you make excuses about what was said, but answer: ‘Evidently, he didn’t know about my other faults, or he wouldn’t have spoken only of the ones he did” - Epictetus
· “This is why I lost my lamps: because a thief was a better than I am at staying awake. But he bought the lamp at a high price. In return he became a thief, he became untrustworthy, he became an animal. This seemed to him a good bargain!” - Epictetus
· “When you see someone often wearing the robe of office, or someone whose name if famous in the Forum, do you not be envious; those things are bought at the cost of one’s life” - Seneca
· “If you look at the effects of anger, and the harm it has done, no plague has been more costly to humankind” - Seneca
· “It is more civilized to laugh at life than to bewail it” - Seneca
· “…to enjoy the sufferings of others is an inhuman pleasure” - Seneca. [Definitely presidential!]
· “Wealth will make you happy, because it will enable you to benefit many; and poverty, as you will then have few things to worry about; and glory, as it will make you honored; and obscurity, for you will then be safe from envy” - Plutarch
· “Toil summons the best men” - Seneca
· “They change their climate, not their disposition, who run beyond sea” - Horace. (He was talking about people traveling to get away from their worries)
· “Sheep don’t throw up their grass to show the shepherd how much they have eaten; after digesting the grass inside, they bear wool and milk outside. So for you too: don’t display your learning to the uninstructed: display the actions that result from the digestion of it” - Epictetus
· “Learning how to live takes a lifetime” - Seneca
· “The study of philosophy is not to be postponed until you have leisure; everything else is to be neglected in order that we may attend to philosophy” – Seneca
Simply delightful.
2. Hard to stay away from 🤖AI, so let’s jump in with a few reports on that. Menlo Ventures dropped their third annual state of Generative AI in the Enterprise report (“menlo_ventures_enterprise_ai_report-2025”) — enterprise AI has exploded from $1.7B to $37B since 2023, now capturing 6% of the ~$300B global SaaS market (page 1), growing ~3x YoY and officially becoming the fastest-scaling software category in history. The application layer grabbed $19B of that spend (page 2), with at least 10 products now generating over $1B ARR (page 3). The build vs. buy debate? Settled — ~75% of AI use cases are now purchased rather than built internally, up from ~50% last year (page 3). AI deals convert at 47% vs. 25% for traditional SaaS (page 4), and 27% of all AI app spend flows through PLG motions (page 4) — Cursor hit $200M revenue before hiring a single enterprise sales rep. Anthropic unseated OpenAI as the enterprise leader, now commanding 40% of enterprise LLM spend (up from 12% in 2023) and 54% of the coding market (page 12-13), while OpenAI cratered to 27% from 50% (page 12). Healthcare leads vertical AI at $1.5B — 43% of all vertical spend (page 10) — because apparently shrinking margins and chronic staffing shortages finally forced health systems to embrace automation. Strip away the hype and only 16% of enterprise deployments qualify as true agents (page 15); the rest are just if-then logic around a model call. Good data. Similarly, Wharton dropped a 🥩 meaty report on enterprise Gen AI adoption (“2025-Wharton-GBK-AI-Adoption-Report”) — 46% of decision-makers now use Gen AI daily, a +17pp leap YoY (page 7), with 82% engaging at least weekly. ~12% of decision makers 55 and older have never used GenAI or don’t use it currently (page 22), smaller enterprises adopting more quickly – maybe because they have to (page 25), ChatGPT and MSFT CoPilot dominate but Gemini is fastest riser (page 34), the ROI reckoning has arrived: 72% now formally track Gen AI returns (page 43), and three-quarters already see positive ROI (page 45) — though Tier 1 enterprises with $2B+ revenues are more likely to shrug and say “too early to tell” (34%) while smaller firms pocket the gains (page 45). Meanwhile, 60% of enterprises have appointed a Chief AI Officer (page 66), though over half just stapled the title onto an existing role (page 66). Worth a flip.
3. One of the critical questions today is whether we are in an AI bubble 🛁. This report (“is-it-a-bubble”) from Oak Tree Capital does a good dispassionate job talking through the financial basics: “75% of gains, 80% of profits, and 90% of capex” in S&P is due to AI, “It’s worth reiterating that two of the closest analogs AI seems to have in tech bubble history are aviation and broadcast radio. Both were wrapped in high degrees of uncertainty, and both were hyped with incredibly powerful coordinating narratives. Both were seized on by pure play companies seeking to capitalize on the new game-changing tech, and both were accessible to the retail investors of the day. Both helped inflate a bubble so big that when it burst, in 1929, it left us with the Great Depression.”, the median P/E ratio of the leading tech / telco players in Dec 1999 was 41x; right now that’s 31x, “It is not ok to use debt in a venture where the outcome is pure conjecture”. Summary: It is. [By the way, the Economist in its imitable way summarized it this thus: “From now till 2030 the big tech companies are scheduled to spend $5T on infrastructure for AI; to make it work they will need $650B a year; today we are at about $50B”. ]
4. OpenRouter and a16z dropped a monster empirical study on real-world LLM usage patterns (“State-of-AI.pdf”): Open-source models have clawed their way to roughly 30% market share (page 5), with Chinese OSS models surging from a negligible 1.2% weekly share in late 2024 to nearly 30% in some weeks (page 5). Here’s the part that was surprising: roleplay dominates OSS usage at ~50% of all tokens (page 10), while programming came in second at 15-20% (page 9). Meanwhile, reasoning-optimized models crossed the 50% threshold of all inference (page 12), average prompt tokens quadrupled from 1.5K to over 6K (page 15), and sequence lengths tripled to 5,400+ tokens (page 15) — signaling that agentic, multi-step inference is coming (page 18). Anthropic’s Claude captures 60%+ of programming spend (page 18-19), but the competitive pressure is relentless: no single open model holds more than 25% of OSS tokens anymore (page 7). Asia’s share of global spend doubled from 13% to 31% (page 24). The authors introduce a delightfully named “Cinderella Glass Slipper” effect for retention (page 25) — early cohorts who find workload-model fit stick around like they’ve found their 😍soulmate, while everyone else churns through models like dating apps. Bit hard to grok overall, but good data.
5. A few more macro tech views. Ben Evans does his usual wonderful sweep of tech (“2025_Autumn_AI.pdf”): $400B in 2025 capex from the big four alone (page 17) is roughly equivalent to the entire global telecoms industry, the models themselves keep converging toward commodity status with leaders changing weekly (page 35), a quarter of CIOs have deployed something, but 40% don’t plan anything until at least 2026 (page 57). But here’s the thing: we’ve been here before (page 82): IBM was selling “150 Extra Engineers” in the 1950s with the same breathless promises about automation. And sometimes “automation alone” really is a big deal — UPCs, barcodes, and databases let retailers manage 5x more SKUs after grocery barcodes launched in 1974, from ~10,000 to nearly 50,000 per supermarket (page 59). In the same vein, Accel has a whopper of a report on the AI infrastructure arms race (“Accel - Globalscape”) — ServiceNow, SFDC and Adobe still waiting for the AI bump to their valuations (page 9), ~60% of venture funding in AI is into models (page 21), difference in performance in text LLMs negligible, but still significant for “computer use” and video (page 25), ~$3T in AI revenue needed by 2030 just to break even on the AI data center spend (page 40), equivalent to adding 1-2% to global GDP CAGR (page 41). Meanwhile, AI-native startups hitting ~$6M ARR/FTE vs ~$0.5M for traditional SaaS (page 30) — either they’ve cracked the code or they’re just not sleeping 🥱. I think it’s both.
6. I know it’s not fashionable to talk about anything but AI, but the “traditional” SaaS companies still exist. In that vein, KeyBanc and Sapphire dropped their 16th annual deep-dive into private SaaS benchmarks (“2025_kbcm_sapphire_saas_survey”) — and the vibes have shifted. Median ARR growth bottomed at 15% in 2024 (page 4) before operators started dreaming big again: 20% expected for 2025, 25% for 2026 (page 15). Nearly all respondents are investing in AI for 2025, with 30 of 55 already seeing “material changes” to operations (page 23) — and zero companies classified themselves as “Non-AI” (page 8). The median company is expecting AE productivity to jump from $283K to $454K by 2026 (page 31) while reducing headcount from 13 to 11 AEs (someone’s betting big on agents) and 67% of AI-enabled companies are already monetizing their AI features (page 24). The catch? Private survey respondents have historically overshot their growth forecasts by 4-6 percentage points (page 17) — optimism is a hell of a 💊 drug. Good data, even though it is for mostly smaller companies. Related to that one is High Alpha’s latest benchmarks survey (“2025 SaaS Benchmarks Report”) captures 800+ respondents in the throes of Operation AI — their term for the shift from AI experimentation to operationalization. Those with AI deeply embedded are growing 70% faster in the $1-5M ARR band (page 22), though at a ~5 point gross margin penalty from compute costs (page 23) — a trade most founders would take blindfolded. The efficiency wave is brutal: later-stage companies slashed headcount 42-59% since 2022 (page 30), with engineering taking the biggest AI-driven cuts at 42% of respondents (page 47). The result? Upper quartile ARR per employee at $50M+ ARR now sits at nearly $400k (page 28), matching public SaaS medians (page 15). Yet the monetization gap persists: 70% have shipped AI features, only 41% are charging for them (page 24). Outcome based pricing models have an outsized impact on growth (page 24). Later stage ($50M+ in revenue) companies are operating ~60% leaner than in 2022 (page 30). Companies are measuring AI impact mostly through “informal team feedback” (page 50), and only 11% describe the productivity gains as transformative (page 51). Shocking no one, default in-office firms are growing faster than those predominantly remote (page 52). Two-person founding teams are growing fastest at 41% YoY (page 54).
7. On to a space that I know little about – retail, and specifically the rise of agentic commerce. Morgan Stanley’s inaugural agentic commerce 🛒 tracker (“Internet_Tracking_the_Path_to_~$385bn_of_Agentic_Commerce”) sizes the prize at $190bn-$385bn in GMV by 2030, representing 10-20% of all US e-commerce (page 1). The adoption picture: 45%/32%/22% of Americans used ChatGPT/Gemini/Meta AI last month (page 1), while retailer-specific assistants like Amazon Rufus and Walmart Sparky hover around 10% each (page 2) — meaning the broad platforms are running circles around the incumbents at 2x-6x adoption (page 2). Here’s where it gets interesting: 53% of ChatGPT users are already researching and comparing prices on the platform (page 3), and 36% have made actual purchases (page 4), which implies 16% of all Americans bought something based on ChatGPT’s recommendation last month (page 5). The retail-specific tools? A paltry 3-4% purchase conversion (page 5). The early category winner? Groceries and household goods, with 49%/41% of AI purchasers buying in those categories (page 5), which conveniently happens to be 19% of Morgan Stanley’s 2030 agentic commerce forecast. In a similar vein, another Morgan Stanley report (“2025-11-17-Morgan Stanley-Internet Agentic Shoppers Are Coming”) breaks down the economics for retailers: the ones that make the most from ads have the most to lose from agents (page 28, 29). [Hat tip to Carlos Lacayo for the second one]
8. Speaking of shoppers, this time human ones, New Consumer and Coefficient Capital have a juicy report on consumer trends for 2026 (“Consumer-Trends-2026”): Average wealth in the US is 5x the median (page 21), high income Americans responsible for nearly 50% of all consumer spend (page 22) - no inequality here – clearly a “affordability hoax”, Gen Z & Millennials represent ~40% of the US population (page 34), 14% of Gen Z or Millennials have appeared as a guest or host a podcast – WTF! (page 36), and 75% of the same cohort consider ‘content creator’ a viable career – we’re so fucked (page 41), US consumers spend more than 3x on OnlyFans than NYTimes – see previous comment (page 69, page number checks out!), ~1/4 of all US households have a GLP user – mucho grande (page 80). So, we are fat, lecherous, vain and stupid. Presidential! 🍊🤡
9. Ok, now on to my favorite part –⚒️ tools. This time, not another breathless claim about more AI tools, but rather a few tools in the “meat and potatoes” category. The first is a Mac workspace switcher that allows you to “lock in” apps per workspace (think a messaging space, vs. a dev space, an email space and calendar space etc.), and then lets you switch between the spaces instantaneously using keyboard shortcuts. The second is even more prosaic: If you are tired of carrying around some plastic cards with a barcode on them (I had one like this for the gym), you can use this tool to create a “Apple pass” that you import to the Apple wallet. Neat. Unfortunately, having the pass in a digital wallet doesn’t make one more likely to go to the gym!
10. And finally, a selected set of 🇧🇬Bulgarian proverbs that ring true. Courtesy the priceless source that is Futility’s Closet.
· Hunger sees nothing but bread.
· The clean gets dirty more easily.
· The devil knows everything except where women sharpen their knives.
· The oversaintly saint is not pleasing even unto God.
· Man is ever self-forgiving.
· God does not shave — why should I?
· Do not salt other people’s food.
· Become a sheep and you will see the wolf.
· The smaller saints will be the ruin of God.
· The wife carries her husband on her face; the husband carries his wife on his linen.
· When a wool merchant speaks of sheep he means cloth.
As you can imagine, there’s a large funnel of material that I have to read / play with to get to this curated list. Almost always a bunch end up on the “cutting room floor”. Instead of throwing it all out, I’ve kept the ones that almost made it but didn’t (there’s a much larger list of crap 💩 that doesn’t get this far!): (i) Meritech 2025 market update (“Meritech report 2025 market updates 20251215”): 30% of US unicorns have declining growth, 65% of all new unicorns were based in the US (ii) For those into biking the Copenhagenize Index returns after a six-year hiatus (“COPENHAGENIZE-INDEX2025_Report_LDef.pdf”): Europe dominates (27 of the top 30 cities were from there; of course there are no US cities in the top 30 cities). (iii) Another quirky report is one that breaks down margins by industry. The bulk of companies sit at ~10%, not surprising that monopolies such as toll road operators, stock exchanges and ports stick out at ~50%, cruise lines at ~16% are higher than airlines at ~12%, beverages are at ~29%, dietary supplements at 37%. (The median margin for SW companies is a paltry 3.5%, but the average is a healthier ~30%).
Hope this helps!
p.s.: A bit of humor before we close in the form of non-traditional passport photos. I don’t think I’ll be able to see a normal passport photo 📷 again with a straight face. Merry Christmas 🎄 and best wishes for the new year!
Thanks,
Akash
