https://www.philk.ai/thoughts.html Observations from altitude on AI, data strategy, higher education, and getting things done. Written by Phil Komarny from Crested Butte, Colorado. en-us hello@philk.ai (Phil Komarny) Mon, 07 Sep 2026 12:00:00 +0000 https://www.philk.ai/images/mtlogo.pnghttps://www.philk.ai/thoughts.html https://www.philk.ai/articles/labor-day-2036.html https://www.philk.ai/articles/labor-day-2036.html Mon, 07 Sep 2026 12:00:00 +0000 When machines do the work, Labor Day stops celebrating rest and starts celebrating effort. A Crested Butte hayfield, 10,000 jobs grading robots at $50 to $90 an hour, and what the first Monday of September means in 2036. In short: One family has farmed the valley outside my door for 150 years: first by hand, then by animal, now by machine, next by intelligence. Labor Day celebrates whatever is scarce, and it was built when work was compulsory and rest was the scarce thing; this year, over the holiday weekend itself, a Palo Alto company posted 10,000 jobs grading robot videos at $50 to $90 an hour, no degree required, a day after the BLS counted 162,000 new jobs for August that will never include them. If the abundance crowd is right, the day flips by 2036: rest becomes the default, effort becomes the rare thing, and the first Monday of September turns into a festival of doing something the slow way. https://www.philk.ai/articles/im-only-data.html https://www.philk.ai/articles/im-only-data.html Thu, 20 Aug 2026 12:00:00 +0000 Schoolhouse Rock taught a generation how a bill becomes a law. Nobody ever made that cartoon for data. Told in the first person by a student record that spent one year being useful. In short: A student record narrates its own life: born on an application in Brownsville, boxed as VARCHAR(50), shipped nightly by a file transfer protocol specified in 1971, and split into nine or ten twins across vendor systems that never meet. The one year it was handed to the student, in UT's TEx program, General Chemistry II pass rates went from 29% to 90% on the same data, and the program was still shut down in 2018 when the Institute for Transformational Learning closed. The silo was always the problem; a model can now read the whole person, but not through a wall. https://www.philk.ai/articles/the-gray-twilight.html https://www.philk.ai/articles/the-gray-twilight.html Sat, 15 Aug 2026 12:00:00 +0000 Universities didn't just outsource their data to vendors. They outsourced the work — and with it, the expertise that made them worth accrediting. In short: Higher ed's outsourcing problem was never the data. It was the work. Online program managers take fifty to sixty percent of tuition on contracts of ten years or more, and 2U, the biggest of them, bought edX for $800 million in 2021 and filed for Chapter 11 in 2024 while its partner universities emerged without the capability they had rented. The Grateful Dead gave away the recordings and kept the mailing list; the way back is doing the work in-house, badly at first, in public. https://www.philk.ai/articles/the-ampersand.html https://www.philk.ai/articles/the-ampersand.html Fri, 01 May 2026 12:00:00 +0000 We trained a generation to speak the language of machines. The machines learned the language of people. The bridge people are about to inherit the building. In short: We trained a generation to speak the language of machines, and the machines learned the language of people. At Robots & Pencils, which grew 3,400% in eighteen months, the rare hire was the ampersand: the developer who notices kerning, the designer who reads the API docs. AI closes the gap inside that person between what they can see and what they can produce, and with specialist technical skills carrying a half-life of about two and a half years against roughly twenty-five for a humanities education, the play is one discipline that ages well and one that compounds fast. https://www.philk.ai/articles/noise-as-news.html https://www.philk.ai/articles/noise-as-news.html Fri, 01 May 2026 12:00:00 +0000 Why 'another model dropped' stopped being a story, and what the actual story is. In short: Four major Anthropic drops in roughly thirty days and GPT-5.5 just 49 days after GPT-5.4: the release cadence serves capital markets, not your Monday morning. Current models are already good enough to draft, summarize, evaluate a transfer credit, and run real agentic workflows; the bottleneck is your data and your institutional muscle. The real shift is that coding agents make it economically rational to own your data, workflows, and AI surface area instead of renting them, which is how a team of five ships like thirty. https://www.philk.ai/articles/service-as-software.html https://www.philk.ai/articles/service-as-software.html Wed, 01 Apr 2026 12:00:00 +0000 The 25-year SaaS sales engine is ending. Pay for outcomes, not seats. Humans state intent. Agents deliver it. In short: For twenty-five years SaaS sold access to a tool and left the outcome as your problem. Service as Software inverts it: agents do the labor and invoices follow finished work. One senior engineer took a CRM from proof of concept to production on AWS in a single day with 18 agents and 67 pull requests, at an idle compute cost of zero. In an era of agents your moat is your description of the world and your willingness to pay only for results, and higher ed, the most over-licensed vertical in enterprise software, has about thirty-six months to make peace with that. https://www.philk.ai/articles/markdown-rebellion.html https://www.philk.ai/articles/markdown-rebellion.html Wed, 01 Apr 2026 12:00:00 +0000 The ERP ran on COBOL-adjacent logic written before the internet. The replacement runs on a folder of text files. A hero's journey through the death of the ERP and the birth of the intelligent institution. In short: A university describes itself in plain markdown files (identity, programs, policies, learner profiles), isolates the transactional systems in an execution plane, and lets governed agents read the description instead of querying the ERP. A stopped-out student with 47 credits got a personalized degree map by text in 90 seconds; a $40 million ERP upgrade became a $400,000 annual investment in self-description; a CRM vendor's renewal price dropped 35% overnight once the institution owned its own intelligence. Salesforce's AgentScript is the same idea with a landlord. https://www.philk.ai/articles/higher-ground.html https://www.philk.ai/articles/higher-ground.html Wed, 01 Apr 2026 12:00:00 +0000 Glasswing is the canary. Your data strategy is the only high ground left. In short: Anthropic withheld its Mythos model from the public and gave it to forty defenders instead, after it exploited Firefox vulnerabilities 84% of the time where the previous model managed under 1%. Attackers now move at machine speed, and models score 84 to 92 percent on USMLE Step 1, so both the security posture and the knowledge-transfer business of most institutions are exposed at once. The only defensible position is a clean, connected, ruthlessly maintained data architecture, because a model is only as useful as the context you can feed it. https://www.philk.ai/articles/building-today.html https://www.philk.ai/articles/building-today.html Wed, 01 Apr 2026 12:00:00 +0000 How a morning briefing script built with Claude evolved from a calendar fetch into something that thinks about my day before I do. In short: Five weeks of working with Claude turned a one-line calendar request into /today, a morning briefing that reads my schedule, filters fifteen intel sources across six platforms, scores each source with a seven-day half-life, and opens with a situational report tying the news to the day's meetings. A colleague installed his own copy from a zip file and a ten-minute conversation, with no developer involved. The interface was a terminal and plain English, and the constraint turned out to be the point. https://www.philk.ai/articles/hlc-keynote.html https://www.philk.ai/articles/hlc-keynote.html Sun, 01 Mar 2026 12:00:00 +0000 An army brat who never went to college is about to tell the accreditors how AI changes everything they audit. In short: Accreditors ask one question every ten years: are you teaching what you say you teach? AI turns that answer from a binder into a daily data asset, mapping every course to every skill to every job outcome. Maryville built on that premise for four years: over 600 AI agents created by its own staff, transcripts evaluated in minutes by KreditFlow, and degree paths that bend around a working adult's life, for the 42 million Americans with some college and no degree. https://www.philk.ai/articles/the-three-laws.html https://www.philk.ai/articles/the-three-laws.html Sun, 01 Feb 2026 12:00:00 +0000 In 2016, Henry King shared three laws of equivalence with me. In 2026, AI is proving every one of them right, and higher education is the case study. In short: In 2016 Henry King laid out three laws in the passenger seat of my car: anything that can be digitized will be, anything that can be quantized for distribution will be, and anything that cannot be made equivalent will be made redundant, each ending with "if not by you, then by someone else." Higher education's product is content delivery and credentialing, and AI tutors and stackable credentials have now made both equivalent. What was supposed to take a generation is playing out in semesters. https://www.philk.ai/articles/the-nod.html https://www.philk.ai/articles/the-nod.html Sun, 01 Feb 2026 12:00:00 +0000 Five Sisters of the Sacred Heart taught me more about AI communication than any conference ever has. Empathetic listening is the killer app. In short: Presenting a university's AI strategy to a board that included five Sisters of the Sacred Heart, I translated every concept into their vocabulary: a neural network as the communion of saints, deep learning as contemplative prayer, a pre-trained model as a novice in formation. I waited for the nod before moving on, every time. The hard part of AI adoption is not the technology; it is the translation, and most technologists never learn the other person's language before asking them to learn ours. https://www.philk.ai/articles/the-hidden-language.html https://www.philk.ai/articles/the-hidden-language.html Sun, 01 Feb 2026 12:00:00 +0000 The BLS has 1,016 job codes. AI is making them obsolete. But universities already speak the language that comes next — they just don't know it yet. In short: The fastest-growing tech job category in St. Louis is "business operations specialists, all other," up 122% to 4,878 postings, which is the Bureau of Labor Statistics admitting its 1,016 occupation codes cannot name what AI is doing to work. A Nature Scientific Data study of three million syllabi found the Department of Labor's skills language already embedded in every course description. Universities do not need new programs; they need to decode the catalog they already have, while 65% of employers move to skills-based hiring. https://www.philk.ai/articles/the-amplifier.html https://www.philk.ai/articles/the-amplifier.html Sun, 01 Feb 2026 12:00:00 +0000 AI is the most powerful amplifier ever built. It doesn't care what signal you feed it. That's your job. In short: Knowledge work was factory work with the damage moved from the back to the wrists: musculoskeletal conditions jumped 25% in a single decade. AI can take the mechanical part, and the same technology can take the human out of the room; both paths start with "let AI make the logo." An amplifier does not choose the signal, you do, so decide what you want more of, let digital tasks return to digital, and invest in purpose with the seriousness currently reserved for GPU clusters. https://www.philk.ai/articles/ikigai-interview.html https://www.philk.ai/articles/ikigai-interview.html Sun, 01 Feb 2026 12:00:00 +0000 I found my reason for being at the intersection of four questions. Then I asked Claude the same ones. The answers changed how I think about both of us. In short: Ikigai asks four questions: what you love, what you are good at, what the world needs, and what you can be paid for. I answered them for myself after thirty years in technology, then put the same four to Claude and did not edit the answers for comfort. Its center landed where mine did: closing the gap between what someone knows and what they need to know, in the moment they need it, which is why the collaboration works. https://www.philk.ai/articles/gutenberg-to-gpt.html https://www.philk.ai/articles/gutenberg-to-gpt.html Sun, 01 Feb 2026 12:00:00 +0000 The printing press gave us access. AI gives us capability. That's not the same revolution — and the difference should keep you up at night. In short: Gutenberg's press put knowledge in ordinary hands, but a book never did anything for you; the gap between reading and doing was filled by five centuries of credentialed professions. AI closes that gap, which is how a colleague built in twenty minutes a nonprofit dataset that was going to take her until March. The cost is that the friction was load-bearing, the models are not neutral, and society had two centuries to adapt to print and has perhaps two decades for this. https://www.philk.ai/articles/bone-dry.html https://www.philk.ai/articles/bone-dry.html Sun, 01 Feb 2026 12:00:00 +0000 Colorado's snowpack is at historic lows. Denver is building data centers that drink 805,000 gallons a day. The future of AI isn't bigger — it's local. In short: Colorado's snowpack sat at 52% of median in February 2026, the lowest since statewide records began in 1987, while a single Denver data center campus is being built to use up to 805,000 gallons of water a day for cooling. Centralized AI is the mainframe era again, and it does not scale when the thing it runs on is running out. The alternative already exists: small, task-specific models running locally on your own data, which Gartner expects organizations to use three times more than general-purpose LLMs by 2027. https://www.philk.ai/articles/word-processor-moment.html https://www.philk.ai/articles/word-processor-moment.html Thu, 01 Jan 2026 12:00:00 +0000 AI in higher education isn't a calculator replacing a slide rule. It's a word processor replacing a typewriter, and that changes everything. In short: The calculator replaced the slide rule and nothing structural changed, which is why provosts love the analogy. The word processor made revision free and dissolved the typewriter market; Smith Corona's last machine was "perfect, and perfectly useless," and the company was bankrupt by 1995. AI makes the existing model of higher education optional rather than better, and for the 42 million Americans who started college and never finished, the difference between the two analogies is everything. https://www.philk.ai/articles/why-were-here.html https://www.philk.ai/articles/why-were-here.html Thu, 01 Jan 2026 12:00:00 +0000 AI isn't coming, it's here. The mission behind the mountain: solving real problems, building community, and going vertical on AI. In short: AI is not coming; it is here, and the people who figure it out first are the ones who stop waiting for the manual. Above the Noise is a working shop, not a consultancy deck: automation that gives an ops team its Tuesdays back, integrations that end the copy-paste between four tabs, and a Discord where practitioners show their screens. Written from Crested Butte, Colorado, at 9,000 feet, for people who are ready to build things that work. https://www.philk.ai/articles/the-911-doctrine.html https://www.philk.ai/articles/the-911-doctrine.html Thu, 01 Jan 2026 12:00:00 +0000 Porsche refined the 911 for sixty years. Higher ed named its biggest platform after a blackboard. One approach survives. The other is about to find out. In short: Porsche spent sixty years refining the 911 for the driver: same silhouette, every change earned. Higher education preserved the lecture, the credit hour, and the semester for the institution's convenience, then named its biggest platform Blackboard and scanned the classroom instead of reimagining it. AI is the engineering that makes real refinement possible for the first time, if it is used to re-engineer the chassis rather than repaint the old one. https://www.philk.ai/articles/scaling-your-talents.html https://www.philk.ai/articles/scaling-your-talents.html Thu, 01 Jan 2026 12:00:00 +0000 How AI amplifies what you already know. From $250/month to $2.50 with Claude Code. Digital empowerment, not replacement. In short: My daughter was paying $250 a month for a booking platform. In one Saturday with Claude Code I built a booking system around her actual workflow, and hosting runs about $2.50 a month. I learned nothing new to do it: thirty years of HTML, CSS, and systems knowledge finally had a tool that could keep pace, which is the point. These tools make experience worth more, not less. https://www.philk.ai/articles/culture-eats-process.html https://www.philk.ai/articles/culture-eats-process.html Thu, 01 Jan 2026 12:00:00 +0000 Why your methodology doesn't matter until you fix what's underneath. Culture eats process for breakfast. Data is the great equalizer. In short: Waterfall, Agile, SAFe, Scrum, Kanban: the framework changes and nothing underneath moves, because process cannot fix a trust failure between departments. Shared data is the one lever that breaks silos without asking everyone to hold hands, and it is also the precondition for AI, which is why an AI platform that needs data from fourteen walled-off systems becomes an expensive science project. Fix the foundation first.