2026-09-24
In "Top AI Analyst Unpacks Today's AI Hype Cycle" (Benedict Evans, filed 2026-09-24), Evans argues that debating whether AI is 'bigger than the internet or electricity' is analytically unfalsifiable and unproductive, and that the more useful move is pattern-matching against past platform shifts (mobile, semiconductors, PCs, the internet) while bracketing off the separate binary possibility of true AGI/ASI, which would render ordinary economic questions moot. Drawing on DAU/WAU/MAU usage data, he shows AI's daily active usage today is concentrated among software developers and self-employed knowledge workers, with most people using it only occasionally, a picture he contrasts with hype narratives of an already-transformed economy. He challenges AI-driven job-loss predictions on the grounds that automation historically changes job content without necessarily cutting headcount (citing accountants surviving spreadsheets, partly via Jevons paradox), and argues job-exposure radar charts oversimplify what people actually do (his 'perfect sphere horse' critique). His practical test for automatability is whether enough training data exists and whether outcomes can be verified at scale, which he says explains why coding became LLMs' strongest early product-market fit while judgment-heavy, tacit-knowledge domains lag. He's skeptical any single stack layer, including foundation-model labs, will capture the whole value chain (mobile operators never captured the value of ~1000x traffic growth), doubts OpenAI-as-Windows comparisons given the absence of LLM network effects, and reads a frontier lab that recently 'flunked out then jumped back near the top' of leaderboards as evidence frontier capability is still purchasable rather than a durable moat. He frames current enterprise AI adoption as stuck at shallow co-pilot rollouts and scattered pilots short of real organizational reimagination, a gap he expects consulting and vertical-software firms to converge on filling, and predicts that, as the internet devastated newspapers while leaving heavy machinery mostly untouched, some industries will turn out surprisingly unaffected by AI. As of this filing, many of the piece's concrete referents (specific leaderboard standings, named lab strategies, MCP as the current protocol) are already dated, but its methodological throughline, tracking capital/infrastructure buildout, frontier-lab consolidation, the Pareto curve of use cases, and model-vs-application value capture, remains the durable takeaway.
2026-09-24
On 2026-09-24, a conversation between Austin Vance (Focused) and Harrison Chase (LangChain), 'Agents in the Enterprise: Deep Agents, Evals, and Ambient UX,' argued that enterprise agent-building is now its own discipline distinct from vibe-coding a demo: evals have become the new product spec, expanding as fast as the agent itself evolves and forcing new roles like an 'AI reliability engineer'; the highest-value enterprise UX increasingly deletes the chatbot in favor of ambient agents embedded in inbox, ticketing, and Slack workflows with humans kept at approval checkpoints — LangChain's own Engine (a background LangSmith agent that scans traces every six hours and proposes fixes) only shipped once PR-opening was handed back to a human click, and its Slack-native coding agent Open SWE reportedly doubled internal doc PRs simply by removing repo friction; observability tooling built to cut hallucinations is doing double duty as audit and governance infrastructure, reportedly letting a bank data scientist breeze through a scheduled model-risk review using LangSmith traces and evals; and rising eval costs (with Uber's CFO cited as saying the company burned through a year's coding-agent token budget in four months) are pushing teams toward a tiered eval-testing pyramid and toward staying model-agnostic across the model, harness, and memory layers (LangChain's flexible Deep Agents vs. its deterministic LangGraph) rather than building on any one frontier lab's stack.
2026-09-24
2026-09-24 — On the Kajabi Unplugged podcast episode "Everyone's Excited About AI but Nobody Can Keep Up," returning co-founders Kenny Rueter and Travis Rosser argue that AI's real danger is confidence rather than error rate: a friend's Claude-generated supplement recommendation for AFib could have killed him were a doctor not involved, and a guitar-teacher Hero reports AI getting high-level music instruction wrong with total certainty, prompting their Twain-paraphrase framing that "it's not what you don't know that gets you in trouble, it's what you know for sure that just ain't so." They tie this to a broader glut of detectably "soulless" AI content on X and LinkedIn and to Tim Ferriss's self-tracked book-sales collapse — flat in 2022, then -6%, -14%, over -40% in 2025, and -57% year-to-date in 2026 since ChatGPT's launch — as evidence that markets for undifferentiated how-to information are being wiped out, while every $1M+ Kajabi Hero they interviewed retains a direct human component (1:1 work, community, or coaching) and uses AI only as an amplifier, never a replacement, summarized as "AI can give me information. AI cannot give me an outcome." Their resulting advice to new creators is deliberately minimalist — one channel, one offer, a simple consistent cadence — rather than chasing AI's 24/7 omnipresence, illustrated by a Hero running a weekly free pull-up challenge with almost none of Kajabi's feature set.
2026-09-24
In a 2026 interview, Ramp CPO Geoff Charles (transcribed as 'Jeff') argues that AI agents like Claude Code have pushed the cost of building software toward zero, so Ramp gives every function — not just engineering — unrestricted access to tools such as Claude Code, its own Inspect agent, and a Voice-of-Customer research agent, letting PMs, designers, and even salespeople ship real production code and documentation directly rather than routing work through engineers or writing specs for humans to build. Because building is no longer the bottleneck, Ramp has compressed planning cycles from roughly three years to about three months, auto-approves a double-digit share of pull requests by assessed complexity, and folds release reporting, sales enablement, and help-center docs into an automated release pipeline; leadership's job has shifted from directing individual execution to root-causing systemic failures in a shared prompt, skill, or design-system component so fixes generalize instead of repeating as one-off feedback. Charles contends the traditional PM value proposition — writing the best spec, managing stakeholders, understanding customers — is now obsolete, and the role is bifurcating into hands-on builders who ship inside the product and GM-style strategists who own positioning, competition, and monetization; Ramp operationalizes this via a formal L0-L3 AI-proficiency ladder, an AI-fluency hiring bar including a build-a-product interview stage, and internal token-usage tracking to see who is keeping pace. He frames the underlying bet — paying an AI agent more in tokens than a salaried employee's pay if it does 10x the work — as exploiting a limited competitive window before AI can 'one-shot' a full platform like Ramp's, and predicts the end state of AI-native software is agents doing users' jobs rather than UIs for users to operate themselves, so requiring a login may itself become a design failure. As a first-person account from a well-funded company with clear incentive to showcase its own transformation, and with Charles himself noting most of what he demoed was built in only the prior few months, the specific efficiency and adoption claims here should be read as self-reported and likely to age quickly.
2026-09-24
On 2026-09-24, filed from Dan Shipper's (Every) podcast forecast "AI predictions: Job markets, Codex beats Claude, and the death of org charts": Shipper argues the AI job apocalypse won't happen, pointing to Every's own headcount doubling despite maximal AI adoption, and instead predicts work bifurcating into everyone having at least one personal delegate agent while most professional work migrates into an agentic "operating system" (Codex, Claude Cowork) that becomes the primary work surface; he argues SaaS survives but must restructure around agent-friendly interfaces (CLIs, parseable UIs, live-reflected actions) since agents multiply effective users rather than eliminating them, and notes that the CLI-centric work style Every itself "speed ran" is already giving way to GUI-wrapped agent environments within about a year, evidence that any given "AI-native interface" is a moving target rather than an endpoint. He contends benchmarks (including the cited METR autonomous-task-duration measure) don't map cleanly onto job replacement because expert work is mostly about framing problems rather than solving pre-framed ones, and that as non-technical staff produce more technical output (citing one example of roughly 50,000 Codex instances triaging pull requests down to about 1,000 merges) the organizational bottleneck shifts from production to integration/curation, which increases rather than decreases demand for engineers; PMs and full-stack designers are named the biggest near-term winners because AI collapses the gap between taste and execution. Organizationally he now favors a single shared, actively-maintained company "super agent" (citing Shopify's River) over per-employee personal agents, walking back an earlier "agent for everyone" view, and insists every agent still needs a dedicated human (a "forward-deployed engineer") to stay useful, captured in his line "automation is a lie." The episode explicitly frames these as provisional predictions to be scored against reality a year later.
2026-09-24
A September 2026 YouTube interview with Stripe's Steve Klaliski describes how Stripe now ships roughly 1,300 PRs a week with no human coding beyond review, produced via internal 'minions' — Slack-triggered, cloud-hosted coding agents (built on a Goose-based harness with MCP access to Stripe's docs, code search, and CI) that one-shot resolve a prompt using the same developer-experience infrastructure (docs, blessed paths, CI, blue-green rollback) that already made human engineers productive, so that, in his words, 'what's good for the developer is good for the agent.' He argues cloud dev environments, not local laptops, are now the hard ceiling on how many agents can run in parallel, that non-engineers (PMs, designers) are already triggering minions with plain-language prompts, and that as coordination and code-writing costs collapse the bottleneck in development shifts to review and to generating enough good ideas. The most forward-looking claim is that Stripe is extending this stack so agents become economic actors: a birthday-party demo used a Machine Payment Protocol co-designed with Tempo to have an agent pay Browserbase, Parallel AI, and Postal Form on the fly (about $547 total, itemized in an 'agent receipt,' plus a Stripe Climate offset for the compute), previewing what Klaliski calls a coming 'agent economy' in which token costs and real-currency costs converge. The human-agent collaboration techniques he describes — asking agents to explain/justify themselves, 'breadcrumbing' via git diffs, and codifying one-off approaches into reusable skills — are offered as mirroring, uncertainly, how he guides his young children. The interview is promotional in tone and its headline PR-volume and cost figures are unverified, but it offers a detailed map of Stripe's current agentic-engineering stack.
2026-09-24
In "Why companies are becoming a series of loops" (a16z, YouTube), Anish Acharya argues against the "permanent underclass" fear of AI-driven unemployment as a "dark fantasy," pointing to unusually distributed opportunity (roughly 20 competitors at every layer of the AI stack instead of one winner), still-strong hiring in supposedly "cooked" fields like radiology, and grassroots adoption even among non-technical staff — exemplified by Kavak's "Jedi Academy," where mechanics ship production AI agents after a six-week course. He frames company-building as a progression from prompts to agents to cascading "loops" that climb to a local maximum and then plateau, requiring human intuition to find the next hill to climb; as loops absorb routine execution, human differentiation shifts toward ambition-setting, a shift he says has already changed a16z's own investment criteria from rejecting overly ambitious pitches to avoiding ideas that are too small. He rejects fast-takeoff/RSI fears in favor of "autocatalytic effects" (new tech improving an existing process rather than true recursive self-improvement) and an electricity-diffusion analogy (roughly 40 years to reorganize factories around electricity) to argue economic diffusion of AI will lag model progress and blunt any runaway effect, and contends many industries aren't "intelligence bound" so their competitive equilibria will mostly survive AI intact. On tactics, he frames model-tier choice as a Pareto-efficiency decision — cheap, RL-tuned open-weight models for bounded-upside functions, frontier models for functions with unbounded or hard-to-calculate upside, since even a routine support call might be the first thread to a major insight — and offers an "infinite-intelligence, near-zero-cost" reorg thought experiment for leadership teams. His biggest flagged opportunity is product design for connection and happiness rather than productivity ("people want to spend time, not save it"), alongside claims that moats are "discovered, not designed" through shipping (Cursor, Granola), that deliberately expensive "software Birkin bag" pricing can signal fit, and that organic word-of-mouth distribution has replaced network-effect bootstrapping now that major platforms defensively block others from building on top of their networks. He closes with personal-practice advice: adopt an "AI-first" reflex before starting any task, maintain a low-stakes "chassis" project to build model intuition by shipping weekly, and notes that AI skeptics typically convert after one genuine moment of AI-made joy.
2026-09-23
On 2026-09-24, a conversation with musician-technologist Jack Conte (posted to YouTube as "A Conversation with Jack Conte") argued that new creative mediums are always first used to imitate the medium they replace — early cinema mimicking theater's wide shots and overacting, early sound recordings mimicking a live band played straight through, and today's AI generative tools reproducing hand-drawn, filmed, or human-performed aesthetics — before artists discover the new medium's own native language, pointing to Méliès's use of the film cut and Revolver-era studio techniques (tape loops, backwards playback, Leslie speakers) as the pivot moments in the prior two cycles and arguing what people dismiss as AI "slop" is simply this same early imitative phase; grounding the argument in economic case studies (Vitaphone paying for itself within months for theater owners while ~22,000 musicians lost work, the Novachord-driven shift of orchestras from scoring 100% of #1 hits in 1941 to essentially none today via Baumol's cost disease, and his own Pomplamoose business model repeatedly broken and rebuilt by YouTube algorithm changes, streaming, COVID, and TikTok) he insists creators must be paid for work used to train AI models — noting AI companies' own paid deals with Disney, Condé Nast, Vox, and Warner Music undercut their fair-use defense — while distinguishing survivable economic disruption from creators' more painful loss of purpose, and closing by arguing genuine art requires risk and a specific human identity (Elliott Smith, Kendrick Lamar) that pattern-matched AI output has not yet shown it can autonomously invent.
2026-09-23
2026-09-23 — Jenny Wen, head of design for Claude Cowork at Anthropic, argues in a talk titled "The design process is dead. Here's what's replacing it." that the canonical design process — research, diverge, converge, "trust the process" — is effectively dead on AI-native teams, not because designers chose to abandon it but because engineers running many concurrent AI coding agents now ship faster than that process can accommodate; design work has consequently split into execution-support (mocking time down from roughly 60–70% to 30–40% of her time, replaced by "jamming" and last-mile implementation directly in code) and much shorter 3–6 month vision-setting instead of multi-year roadmaps. She justifies skipping exhaustive up-front mockups on the grounds that nondeterministic AI models can't be captured in a static clickable prototype, so shipping early under an explicit "research preview" label with stated flaws and then iterating visibly and fast is what actually builds user trust, while a human still has to remain the accountable decider of what gets built, since AI can weigh in on disputes between people but can't resolve them. She also gives concrete organizational color from Anthropic: Cowork's viral "built in 10 days" framing masks years of separately-prototyped interaction patterns assembled at the end; the ambiguous internal "Cloud Studio" prototype seeded both the Claude skills markdown-instruction framework and several Cowork UI patterns after she scouted it VC-style using Evan Tana's founder/idea legibility grid; her hiring now favors block-shaped generalists, deep T-shaped specialists, and humble "craft new grad" juniors over conventional experience-weighted hiring; and her leadership pairs Radical-Candor-style psychological safety (gauged informally by whether reports feel safe "roasting" her habits) with deliberately doing nominally low-leverage tasks like dogfooding and filing bugs herself to build trust and product fluency. She flags her own views as unsettled — "ask me in 3 months and it might change again" — which is also the crux of the pushback a large share of her audience gave to the headline "process is dead" framing.
2026-09-23
In this Anthropic CPO interview, Mike Krieger argues that AI now handles the bulk of execution at Anthropic itself — he cites 90% of code and over 70% of pull requests as Claude Code-generated — which has repeatedly outpaced supporting infrastructure, forcing rebuilds of the merge queue and PR review process and revealing that the real constraint on AI-driven productivity is organizational rather than a model-capability ceiling; his own skepticism about models having independent judgment broke down not from a benchmark but from Opus 4 giving him genuinely non-"anodyne" strategy feedback, which he pairs with Dario Amodei's SWE-Bench trajectory prediction tracking close to reality as corroborating evidence that pace is exceeding expectations. He frames Anthropic's product moat as arising from embedding product people directly in post-training work with researchers rather than from building features atop a finished model, and argues the company's winning strategy against ChatGPT's consumer dominance is not to chase that hit but to double down on its developer/builder identity, invest in MCP as open shared infrastructure since Anthropic's business is models rather than integrations, and keep product teams focused on comprehensibility, strategy, and evangelism as the "overhang" between model capability and everyday usage grows. Reflecting on Artifact's shutdown, he attributes its failure to classic startup causes — lack of organic traction, the founders' own ethical limits on growth-hacking, and being fully remote making crisis pivots harder — rather than anything AI-specific, and he explicitly warns against re-importing Web 2.0 engagement metrics like time-spent or message count into evaluating Claude, since optimizing for likability risks sycophancy instead of measuring whether Claude actually helps people finish work.
2026-09-23
On the 2026 creator-economy M&A wave, RockWater founder Chris Erwin frames 2026 as the year the space became an active hunting ground for media companies, agencies, and investors, citing his firm's 2026 predictions report and describing the market as entering the 'sophomore year' of its growth cycle -- excitement in 2021, a pullback from 2022-2024, resurgence in 2025, and now industrialization in 2026 with more capital and more sophisticated buyer types (traditional entertainment, digital agencies, SaaS, AI companies) converging on three deal zones: talent representation, digital ad agencies, and social publishers. Erwin's central claim, discussed on the podcast 'Buying Time: Creator Economy M&A,' is that buyers in this immature market are purchasing time rather than cash flow -- real capabilities, high-quality teams, and client relationships that would take 3-5 years to build organically but can be acquired and market-ready in roughly six months -- illustrated across three traditional-media buyer archetypes: publishers like People Inc (which RockWater advised on its Feedfeed purchase) and Future plc (She Likes) buying audience access and UGC licensing capability; studios like Fox, flush with cash from selling its studio business at the peak, assembling a multi-front creator-economy operating structure (Red Seat Ventures, The Lighthouse, Holy Water, Meet Cute, Fox Creator Studios, a Chain equity stake) to learn how to work with modern talent; and ad agencies like WPP and Publicis rolling up creator-marketing infrastructure (Publicis alone acquiring four to five agencies including Captivate, Influential, and BR Media Group in two years) to replace linear-TV ad sales. The conversation locates the main integration risk in preserving a target's editorial and cultural authenticity, contrasting slower traditional-media integrations with the creator-to-creator acquisition of some of Mark Rober's YouTube channels by Overlap, which Erwin expects to integrate faster given shared operating culture, and closes with Erwin's buyer posture (still underwrite fundamentals, have an articulable thesis, stay flexible, favor boldness over prolonged diligence) and career advice (find leadership and mentors who care about your growth, and specialize in what you're already good at rather than chase passion).
2026-09-23
Freshworks CPO Shini Ragaban argues that AI has killed the traditional linear PM→UX designer→engineer handoff, merging those roles into a single "product builder," and he demonstrates this live via Freshworks' internal "AI PDLC" pipeline (Cursor, PRD Genie, Figma Make, Agent Studio, and a Claude-MCP connector) that he says took the company from a six-month release cycle to a two-week one. The live demo itself surfaced a nontrivial catch: the AI-generated PRD flagged that employee experience (EX) uses a different "service catalog" concept than customer experience (CX), showing that grounding in company-specific data can catch domain nuances a naive reuse of an existing feature would miss. Prototyping in Figma Make was explicitly not one-shot: the tool initially misused design-system components, broke on narrow monitors, and failed to render the Sankey chart correctly — the "judgment" value-add the speaker describes concretely consisted of catching these specific gaps through iterative prompting, giving a tangible referent for "co-pilot, not autopilot." The governance mechanism for trust at enterprise scale isn't described as better AI capability but as explicit provenance: every generated markdown artifact cites which reference/data version it drew from, directly answering the host's raised concern about stale or ambiguously-defined metrics being silently baked into an AI-drafted PRD. The organizational effect claimed (PM:engineer ratio compressing from ~1:20 to ~1:1, sometimes with zero dedicated UX designer) is presented as a concrete before/after data point tied to an 18-month timeline, not a hypothetical projection, directly linking the tooling narrative to a headcount/structure claim. Interview practice at Freshworks was changed to match the skill now being valued: candidates are asked to open Cursor live and show their git repository rather than answer behavioral questions, operationalizing the "judgment and curiosity over credentials" argument at the hiring level rather than leaving it abstract. An audience comment flags a labeling slip in the source itself (the chart shown was a treemap, not a Sankey diagram, despite being called one throughout) and separately argues the video only shows the surface: the harder groundwork of building the design system, feeding tech-stack context, and initializing project context is not revealed.
[Source](https://www.youtube.com/watch?v=0qNZVlW8IR4)
2026-09-23
Stripe and Elopage (a Berlin-based creator/digital-entrepreneur platform that just closed a €32M Series A) argue the European creator economy is a structurally growing market rather than a pandemic-era blip, whose real size is hidden because public data only captures the top ~1% of earners, and that serving it profitably requires product-led self-serve onboarding, borderless/localized payments used as a growth lever, and treating creators as "mini startups" rather than hobbyists. The 2018 anecdote where investors called the market 'zero' despite Elopage already having live transaction volume shows how blind institutional capital was to the pre-hype creator economy. Elopage began as a Venmo-style consumer payments app for the German market, failed to monetize app traffic, and pivoted only after digital-nomad users asked to use it for selling ebooks — the product was discovered by its users, not designed top-down. The ~$100/month cost floor to start a creator business, combined with global freelance access via platforms like Upwork, is cited as what makes today's low barrier to entry for creator businesses possible. Digital products are framed as more cross-border-friendly than physical goods specifically because of simpler tax treatment, which the speakers say is accelerating international creator commerce. Creators actually use split-payment/partner-commissioning features, but typically only with 10-20 collaborators at a time — a concrete ceiling on how 'networked' creator collaboration gets in practice, despite the rhetoric of constant cross-promotion. Mark's own guess of creator minimum income (~$1,000/month) undershot Elopage's real average cart size alone (~$270-300 per sale), illustrating how even industry insiders underestimate top-end creator earnings. Stripe frames its own onboarding philosophy ('the product is our number one salesperson') as directly transferable to Elopage's product-led approach, drawing an explicit parallel between a payments infrastructure company and a creator-economy platform.
[Source](https://www.youtube.com/watch?v=hpdsFcdMqCU)