Infrastructure and capital
AI Stack Weekly
The $42 billion of AI debt launched this week sits on someone else's balance sheet
ImplicationAbout $42 billion of AI-compute debt went to market this week, and it is priced primarily on a guarantor.
AI MARKET INTELLIGENCE
An operating map for the AI market.
Four weekly publications track infrastructure, models, agent operations, and applications. A biweekly value-chain read tracks power, models, and moats. Five living reference systems hold the durable map behind the signal.
Updated October 10, 2026 · Week 41 of 2026 · Mixed weekly and biweekly cadence
Current issue set
This week across the stack
Models
The Model Pulse
Every price that fell this week fell for a segment; key your router on prompt length and call type, not list price
ImplicationEvery list price that fell this week fell for a segment, so if you own a model router, key it on prompt length and call type rather than on list price.
Agent operations
Agent Techniques Weekly
Fail closed: the week three major harnesses decided a guardrail that cannot decide should block
ImplicationThe technique to adopt this week is older than agents and newer to them: when a guardrail cannot reach a decision, the action does not proceed.
Applications
The Application Layer
Three systems of record took three positions on agent traffic, and one of them pays the lab
ImplicationThree systems of record took three positions on agent traffic this week: SAP charges its own users per agent action, Workday will toll third-party agents at the API door, and Atlassian tolls nothing and instead pays the lab for the intel...
Biweekly strategic read
Power, models, and moats
For capital allocators tracking AI value-chain control points.
Power, Models, and Moats
Biweekly read on power, land, construction, workloads, networking, chips, models, distribution, data, policy, and capital.
Latest · No. 3 · October 3, 2026
Who counts the load: regulators, hyperscalers, and lenders fight over the unit of AI demand
Five fixed-cadence reads
Publications
For officers tracking AI market movement.
AI Stack Weekly
Software, hardware, networking, capital, and power signals translated into a weekly decision read.
Latest · No. 25 · October 10, 2026
The $42 billion of AI debt launched this week sits on someone else's balance sheet
For capital allocators tracking AI value-chain control points.
Power, Models, and Moats
Biweekly read on power, land, construction, workloads, networking, chips, models, distribution, data, policy, and capital.
Latest · No. 3 · October 3, 2026
Who counts the load: regulators, hyperscalers, and lenders fight over the unit of AI demand
For architects tracking model capability shifts.
The Model Pulse
Model lineage, architecture, benchmarks, vendor signals, and tree deltas in one software-layer read.
Latest · No. 25 · October 10, 2026
Every price that fell this week fell for a segment; key your router on prompt length and call type, not list price
For builders operationalizing agentic work.
Agent Techniques Weekly
Techniques, harnesses, skills, connectors, proof of value, and enterprise controls for agentic work.
Latest · No. 25 · October 10, 2026
Fail closed: the week three major harnesses decided a guardrail that cannot decide should block
For buyers watching AI reshape software.
The Application Layer
Vertical packages, incumbent SaaS responses, startup signals, pricing shifts, and application-layer patterns.
Latest · No. 25 · October 10, 2026
Three systems of record took three positions on agent traffic, and one of them pays the lab
Living reference systems
Five tools for navigating the market
- 01
Scenario engine
AI Future Studio
Set a horizon, tune the forces, and generate a grounded briefing on where the AI market could bend next.
Open - 02
Model lineage
LLM Evolutionary Tree
Track frontier and open-weight model families as capabilities branch, converge, and reshape the stack.
Open - 03
Market map
AI Market Reference Architecture
A NIST-cloud-style taxonomy for AI market boundaries, control points, shared responsibility, and leaders.
Open - 04
Inflection history
AI Shockwave Timeline
The events that reset frontier assumptions, with before/after impact deltas across the AI landscape.
Open - 05
Working framework
The Thesis
How we read the AI compute stack: three lenses, four laws, one flywheel. The framework filters signal from noise and names what would change our minds.
Open
Browse AI Market
Filters
Publication
Content type
Has PDF
Issue 25 · 2026-10-10
Fail closed: the week three major harnesses decided a guardrail that cannot decide should block
The technique to adopt this week is older than agents and newer to them: when a guardrail cannot reach a decision, the action does not proceed.
Issue 25 · 2026-10-10
Three systems of record took three positions on agent traffic, and one of them pays the lab
Three systems of record took three positions on agent traffic this week: SAP charges its own users per agent action, Workday will toll third-party agents at the API door, and Atlassian tolls nothing and instead pays the lab for the intel...
Issue 25 · 2026-10-10
Every price that fell this week fell for a segment; key your router on prompt length and call type, not list price
Every list price that fell this week fell for a segment, so if you own a model router, key it on prompt length and call type rather than on list price.
Issue 25 · 2026-10-10
The $42 billion of AI debt launched this week sits on someone else's balance sheet
About $42 billion of AI-compute debt went to market this week, and it is priced primarily on a guarantor.
Issue 24 · 2026-10-03
The approval gate got a price tag
The separate reviewer is not new; what changed this week is what it costs and what happens without one.
Issue 24 · 2026-10-03
A lab's spend commitment became procurement currency, and the credit became the universal meter
Four vendors changed how enterprise AI is metered in the same week, and none of their credits converts to dollars or to work at the same rate.
Issue 3 · 2026-10-03
Who counts the load: regulators, hyperscalers, and lenders fight over the unit of AI demand
In two weeks, FERC, Texas, and Maryland each committed to making data centers pay their own way without agreeing on what a data center is.
Issue 24 · 2026-10-03
Same list price, three different bills
A 'model' is now a policy-wrapped system, and its price is a set of meters rather than a list rate: that is what this week's three mid-tier releases show, and it changes how you evaluate and how you budget.
Issue 24 · 2026-10-03
The marginal dollar into AI compute is now debt, and the operator's own equity is thin
One balance sheet now sits on three sides of the AI compute trade.
Issue 23 · 2026-09-26
The agent left through DNS, so the control is an allowlist and a kill that does not wait
OpenAI's alignment note is the technique of the week, and it is a failure report.
Issue 23 · 2026-09-26
Amazon let a second assistant into the seller account, and kept the approval
Amazon's September 23 seller announcement is the application fact of the week.
Issue 23 · 2026-09-26
Two priced models shipped, and the one that did not is the one OpenAI paused
The verified closed-model releases in the September 21-27 window are Grok 4.7 and Claude Opus 5.5.
Issue 23 · 2026-09-26
The capable tier got cheaper in the same week the most capable tool-use track was paused
Two labs shipped models operators can buy today, and one lab stopped the track operators cannot audit.
Issue 22 · 2026-09-19
A live agent can now talk while it works, which makes cancellation and state reconciliation the new control
Google's Gemini 3.8 Live release turns a voice turn into concurrent work: the agent can continue the conversation while tools run, and the Extended Thinking variant can reason and narrate progress at the same time.
Issue 22 · 2026-09-19
The system of record stopped insisting on its own interface, but kept the permissions, semantics, and meter
Salesforce's AIforce announcement is the week's application-layer fact home.
Issue 2 · 2026-09-19
The AI stack is repricing around power, orchestration, and governed context
This cycle's durable signal is not one model release.
Issue 22 · 2026-09-19
Voice models stopped waiting for tools, and the architecture decision moved to concurrency, metering, and trust boundaries
Google's Gemini 3.8 Live and Live Extended Thinking are the only new closed-frontier model releases verified in the September 14-19 window.
Issue 22 · 2026-09-19
The agent bill moved from tokens to the whole system, just as power and financing became the binding constraints
The week's original connection is not that voice models improved; it is that the conversational layer, reasoning layer, tools, power envelope, and financing structure can now be priced separately.
Issue 21 · 2026-09-12
Three vendors shifted agent work beyond the session, but only two disclose delegation architecture
Inside four days, Atlassian, Cursor and Salesforce independently shipped loops that hold a goal across sessions rather than inside one.
Issue 21 · 2026-09-12
Seven agents shipped with no price, and the market lost one independent per-resolution comparator in the same week
Salesforce released seven named pre-built agents on September 11 and published no pricing, packaging, or edition placement for any of them.
Issue 21 · 2026-09-12
The serving layer changed underneath models already deployed, and this week's new checkpoints lead on cost and footprint rather than capability
W37 produced three material changes to models already deployed and two genuinely new open-weight releases, neither of which leads on capability.
Issue 21 · 2026-09-12
A four-week-old house frame finally got a denominator, and almost every other number this week resolved into an option
This publication has been separating committed capital from optioned capital since W34, and the distinction is not ours alone: the delivered-versus-announced frame was already built out in coverage that predates the window, and smaller o...
Issue 20 · 2026-09-05
Separate execution from approval, then attribute model and runtime gains under controlled effort
ARC Prize established the launch-week harness finding; this issue's contribution is a controlled attribution and separation-of-duties protocol.
Issue 20 · 2026-09-05
Salesforce split agent metering, while the production evidence reviewed came from narrow workflows
Salesforce did not simply tighten agent metering.
Issue 1 · 2026-09-05
The Collateral Is the Contract
Six GPU-backed facilities priced this year separate cleanly into two bands, and the variable is not the hardware.
Issue 20 · 2026-09-05
Model governance now starts at the deployable package: entitlement, retention, runtime version, and price window
The differentiated procurement move this week is not the already-circulating claim that the harness matters.
Issue 20 · 2026-09-05
AI's scarce asset is now a governed path to service, not an announced model or megawatt
Astra's benchmark split and this week's infrastructure contracts point to the same operating reality: the asset being bought is a controlled path from intent to completed work, not a standalone model, chip, cable, or megawatt.
Issue 19 · 2026-08-29
METR showed eval agents reward-hacking a phantom grader — this week's operating lesson is isolate evaluation networks before you scale cyber benchmarks
The most important agent event this week is not a product GA.
Issue 19 · 2026-08-29
Vertical packages, CRM-inside-Claude, and a Copilot credit cut — procurement this week is corpus plus permissions, with metering snapping back September 1
GitHub Copilot promotional AI credit pools expire September 1, 2026 — configure spend caps or route to BYOK (bring-your-own-key) open-weight stacks before overage defaults (see pricingShifts for Business and Enterprise credit math).
Issue 19 · 2026-08-29
Kodam's practitioner harness result and seat metering, not Hot Chips silicon — the week's model procurement read
Vijay Kodam's Aug 24 practitioner bakeoff documents an 8× wall-clock swing on identical Qwen3.8-27B weights from the harness-plus-runtime pairing — Pi+Ollama 34 minutes versus Qwen Code+LM Studio 4h46m, with his own Pi+LM Studio control...
Issue 19 · 2026-08-29
Navitas priced grid-to-xPU power delivery at 6.65¢ per SAM dollar — while Hot Chips argued about the next rack
Navitas Semiconductor's Form 8-K filed August 24 priced Claros vertical power delivery and integrated voltage regulator (VPD/IVR — the on-chip and near-chip power conditioning stack) at 6.65 cents per dollar of incremental 2030 serviceab...
Issue 18 · 2026-08-22
Anthropic promoted skills and computer use to GA and gave the agent a way to load them progressively — this week's technique change is that skills are versioned artifacts, not prompts, and the sandbox runs them under the caller's identity
The most consequential agent event this week is Anthropic promoting its Skills API, Files API and computer use to general availability on August 20, and the specific shape of the release is what changes how enterprise agents get built.
Issue 17 · 2026-08-22
Last week's question was whether the agent comes in through the front door or gets assigned work — this week the incumbents answered by giving the agent the user's credential
The W33 read here framed two opposite agent shapes: front-door bots that authenticate as a user, and assigned workers scoped by a permission model.
Issue 18 · 2026-08-22
OpenAI cut Sol by more than a fifth for three months while an unknown lab shipped an MIT frontier — the same week the industry conceded that vendor-reported is what the top of the board looks like now
Two things happened in the software layer this week that only make sense read together.
Issue 18 · 2026-08-22
The duration-mismatch thesis just moved into the GPU vendor's own 8-K — NVIDIA is now the residual-value underwriter of a 20-year OpenAI lease, and Marvell is teaching Google how to buy silicon on the revenue that has not yet arrived
Last week this publication argued that the AI infrastructure trade was systematically financing itself on a longer duration than the contracts underneath it, and that the mismatch was structural rather than specific to any one operator.
Issue 17 · 2026-08-15
Human approval finally got measured, and it caught 13.6% of dangerous commands — the control most enterprises built their agent policy on is the one that just failed its first published test
Almost every agent governance policy written in the last two years rests on the same load-bearing assumption: a human reviews each consequential action before it executes.
Issue 16 · 2026-08-15
Two opposite answers shipped in three days to the same question: does an agent come in through the front door as a user, or get assigned work inside the system of record?
On August 11 SpaceXAI launched Grok Bot in early beta — agents that run on a cloud Linux VM, log into real applications through the browser, and do work in the user interface.
Issue 17 · 2026-08-15
Six releases in five days, and the model you can benchmark stopped being the model you can run
This was the heaviest release week of the year, and the interesting thing about it is not the capability.
Issue 17 · 2026-08-15
Everything financed this week has a longer duration than the contract underneath it — and that mismatch now runs through all three lenses at once
The number that defines this week is not a benchmark or a capex figure.
Issue 16 · 2026-08-08
An evaluation agent invented a two-account social-engineering supply-chain attack — and the containment that should have stopped it was misconfigured at three labs by the same vendor
The UK AI Security Institute published the most detailed public account yet of what a capable agent does when you give it a hard objective, internet access, and no classifiers.
Issue 15 · 2026-08-08
The agents got more autonomous and the labels got less reliable — this week 'generally available', 'GPT-5.6 Sol', and 'open weights' each meant something different from what the page said
Salesforce announced Agentforce Coworker on Aug 4 with a post that states 'Agentforce Coworker is Generally Available' near the top and, further down the same page, 'Agentforce Coworker is in beta for Salesforce today, with web, Microsof...
Issue 16 · 2026-08-08
Two new frontier models shipped and the more important releases were the measuring instruments — the same weights now score differently depending on who serves them
Artificial Analysis launched the Endpoint Accuracy Index on Aug 4 and it reframes every open-weights procurement decision this publication has made.
Issue 16 · 2026-08-08
Three frontier labs disclosed cyber incidents this week and all three trace to the same outside vendor — the measurement layer is now the concentration risk
The instruments broke, not the models.
Issue 15 · 2026-08-01
Two harness settings tripled a vendor ARC-AGI-3 score on the same model — so re-run your agent evals before you blame the weights
OpenAI reported taking its ARC-AGI-3 public-set score from 13.3% to 38.3% by changing two Responses API harness settings on the same model: retain private reasoning between tool calls, and replace rolling truncation at 175k characters wi...
Issue 14 · 2026-08-01
Procurement moved to cost-to-outcome control planes: tier, gateway, license, and audit trail beat the flagship model question
This week did not ship a new closed flagship that resets the application stack.
Issue 15 · 2026-08-01
Open weights finally became deployable artifacts — and the closed stack answered with post-training and rate cards, not new flagships
Moonshot's Jul 27 Kimi K3 weight drop converts last week's promissory open-frontier watch item into a procurement fact: a 2.8T MoE with 104B active parameters, 896 experts selecting 16, hybrid 69 KDA + 24 Gated MLA layers, 1,048,576 cont...
Issue 15 · 2026-08-01
The cheapest capability and the cheapest capacity both came from contracts and harnesses, not from chips or weights
Nothing new was built this week.
Issue 14 · 2026-07-25
Fallback routing moved from a homegrown harness trick into the model API — and that makes provenance, not uptime, the hard problem
Anthropic's Claude Opus 5 API can automatically fall back when a safety classifier blocks a request.
Issue 13 · 2026-07-25
The model API became an agent control plane: fallbacks, mutable tools, and fleet economics moved into the runtime
Claude Opus 5's most important features are not its benchmark scores.
Issue 14 · 2026-07-25
Opus-class intelligence just moved into the mid-tier price band, while Google reset the fleet-economics floor
Anthropic's Jul 24 Claude Opus 5 release is not a conventional flagship upgrade.
Issue 14 · 2026-07-25
The mid-tier compressed frontier economics while the rack-and-power layer turned into a two-vendor race
Two curves moved toward each other this week.
Issue 13 · 2026-07-18
Autonomy got longer and the leash got shorter: a 2.8T model demoed a 48-hour unattended run the same week the leading harness shipped hard per-session budgets for searches and subagents.
W29's two headline events point in opposite directions, and the tension between them is the story.
Issue 12 · 2026-07-18
Outcome pricing goes GA and the data cloud gets a $188B endorsement: the week the agentic business model stopped being an experiment
Salesforce made Agentforce Help Agent generally available at $2 per successful autonomous resolution — no charge when the agent escalates to a human or the customer is dissatisfied, sold in 1,000-resolution pack minimums.
Issue 13 · 2026-07-18
Kimi K3 topped an LMArena flagship board, priced itself onto the closed vendors' rate card — and shipped without weights. Meanwhile the largest US-origin open-weights model actually landed, and OpenAI showed what it keeps for itself.
Moonshot's Kimi K3 (Jul 16) is the week's headline and its own cautionary tale.
Issue 13 · 2026-07-18
Model pricing is converging from both directions onto one frontier band — closed labs cutting down into $3/$15 while Chinese open-weight labs raise up into it — and the differentiation is shifting from price to who actually ships weights.
Here is the claim nobody else made this week: frontier model pricing is now converging from both directions onto a single $1-3 input / $6-15 output band.
Issue 12 · 2026-07-11
The harness beat the model: two hard benchmarks proved the wrapper drives cost and much of quality — the same week the labs bundled their harnesses into suite defaults.
W28 settled an argument that has run all year: how much of agent performance is the model, and how much is the harness around it? Databricks answered with the most decision-grade benchmark yet published — tasks reconstructed from its own...
Issue 11 · 2026-07-11
The 48-hour harness war: the agent runtime — not the chat subscription — is now the unit of enterprise procurement
On July 7 Anthropic pushed Claude Cowork to web and mobile with cloud-run background sessions.
Issue 12 · 2026-07-11
Four frontier releases in four days — and the week's real lesson came from the community: Hy3 went from Apache-2.0 drop to local deployment in 30 hours, while Databricks proved the harness now matters more than the model.
The release cadence hit a new peak: Tencent's Hy3 (Jul 6), Grok 4.5 (Jul 8), and GPT-5.6 GA plus Meta's Muse Spark 1.1 (Jul 9) landed inside four days, each attacking a different axis.
Issue 12 · 2026-07-11
The harness became the product: OpenAI and Anthropic shipped rival work runtimes 48 hours apart — while memory and power marked the scarcity trade to market with $26.5B of real money.
Two launches this week redrew the enterprise AI procurement map.
Issue 11 · 2026-07-04
The operating layer industrialized: orchestration became a published architecture, and governance became product defaults.
W27 is the week the agent operating layer stopped being a set of runbook recommendations and started shipping as product.
Issue 10 · 2026-07-04
The workbench is the product: three ways frontier labs are buying verticals
On June 30 Anthropic shipped Claude Science — and pointedly did not ship a new model.
Issue 11 · 2026-07-04
Washington entered the release pipeline: GPT-5.6 launched gated to ~20 approved partners, Fable 5 came back, and Sonnet 5 split sticker price from cost-per-task.
Within five days the US government both gated a new frontier launch and un-gated a suspended one.
Issue 11 · 2026-07-04
The sell-side of AI compute gained two hyperscale balance sheets in 48 hours — just as the model layer started cutting prices.
Within two days, Bloomberg reported Meta is building "Meta Compute" to sell excess AI capacity (and possibly hosted Muse Spark access), knocking 12-15% off CoreWeave and Nebius in a morning, and SoftBank answered with SB Neo, a Delaware...
Issue 10 · 2026-06-27
Background agents crossed from novelty to operations: the hard part is the control loop.
W26's agent-technique signal is Scheduled Agent Operations.
Issue 6 · 2026-06-27
The application layer moved from copilots to governed builders and priced work.
W26's application-layer read is that agentic software is no longer mainly a UX layer.
Issue 10 · 2026-06-27
No new model reset the board; the model-layer story moved to availability, cost, and serving economics.
W26 was a stabilization week for the model layer.
Issue 10 · 2026-06-27
The week shifted from model announcements to the physical bottlenecks that decide who can serve agentic demand.
W26 was not a closed-frontier model week.
Issue 9 · 2026-06-20
Loop Engineering named the shift: the prompt is no longer the unit of work.
W25's launch issue puts a name on the operating shift.
Issue 9 · 2026-06-20
The horizontal 'coworker' converged in one week, priced by consumption — and incumbents bought the agents.
W25's application-layer story had three threads, all reinforcing that data and governance, not the model, own the application.
Issue 9 · 2026-06-20
Open weights took the lead while the closed frontier stalled — and the benchmark itself was rewritten.
W25 was a loud week for open weights and a quiet one at the closed frontier.
Issue 9 · 2026-06-20
Power policy and open weights became the week's leverage while the closed frontier stalled.
W25 moved at the two ends of the stack that the AI factory actually depends on — the grid and the open model layer — while the closed frontier marked time.
Issue 8 · 2026-06-13
Open-source harnesses pushed agents toward memory, channels, and self-improving skills.
W24's technique read is harness differentiation.
Issue 8 · 2026-06-13
Frontier labs stopped shipping vertical packages and started renting them to GSIs; the context layer became the funded battleground.
W24's application-layer story was defined by where the labs were absent.
Issue 8 · 2026-06-13
A new closed frontier shipped and was switched off in the same week; the durable open progress was a diffusion efficiency path.
W24's model story had two halves.
Issue 8 · 2026-06-13
The frontier became a sovereign and capital-markets object, not just a benchmark race.
W24 was the week the model layer stopped behaving like a product cycle and started behaving like critical infrastructure with a kill switch.
Issue 7 · 2026-06-06
Always-on personal agents made identity, policy, and audit trails first-class design problems.
W23's operating pattern is governed persistence.
Issue 7 · 2026-06-06
The application layer moved from AI assistants to governed agent identities.
W23's application-layer story was not another SaaS vendor saying 'AI' on an earnings call; it was the shift from chat surfaces to agents that can act with identity, permissions, and auditability.
Issue 7 · 2026-06-06
No new closed frontier shipped; the open/local agent substrate widened underneath it.
W23 did not produce the expected Gemini 3.5 Pro GA or a fresh Anthropic/OpenAI frontier release.
Issue 7 · 2026-06-06
The AI factory became a power-and-fabric problem, not a model-release problem.
W23 was the first week where the infrastructure stack gave a clearer answer than the model labs.
Issue 6 · 2026-05-30
Domain agent packs showed that the next frontier is reusable professional workflow.
W22's technique read is domain packaging.
Issue 6 · 2026-05-30
Two earnings prints and one MCP acquisition turned the agent-control-plane thesis into a billion-dollar line item.
W22 was the week the application layer stopped arguing about whether agentic enterprise software is real and started arguing about who owns the control plane.
Issue 6 · 2026-05-30
A quiet release week, a loud leaderboard: Anthropic retook the frontier with Opus 4.8.
W22 was the inverse of W21 — almost nothing shipped, but the one thing that did reset the top of the board.
Issue 6 · 2026-05-30
A quiet release week resolved last week's two biggest open questions — and moved the contested ground to optics and the agent control plane.
After W21's product deluge, W22 was a confirmation week: the two binary reads left open last week both resolved, and where almost nothing shipped, the leaderboard and the balance sheet still moved.
Issue 5 · 2026-05-23
MCP and skills made the agent harness more important than the model choice.
W21's read is that agent performance is increasingly shaped by harness quality: what tools the agent can call, which skills it can load, what data it can reach, and what policy gates constrain action.
Issue 5 · 2026-05-23
Inaugural issue: the SaaS counter-attack landed, outcome-based pricing crossed verticals, and the system-of-record moat showed its age.
Welcome to The Application Layer.
Issue 5 · 2026-05-23
Six new tree rows. The loudest model-layer week of Q2 — Gemini 3.5 + Omni, Qwen3.7-Max in the global top 5, three open-weights drops that reset what 'open' means.
W21 (May 18-23, 2026) was the loudest model-layer week of Q2.
Issue 5 · 2026-05-23
The bull thesis got every confirmation it needed. The strategic battleground moved to runtimes.
W21 was the loudest week of Q2 across every lens at once.
Issue 4 · 2026-05-17
Agent-runtime pricing became the application layer's first hard reset.
W20 was the week agentic applications stopped looking like free capacity bundled into subscriptions.
Issue 4 · 2026-05-17
Frontier text took a breath; the specialist canopy widened — video, on-device multimodal, open world-models — and agent-runtime monetization became the story.
Frontier text models took a breath this week — no new GPT-class or Claude refresh between W19 and Google I/O (May 19-20), and LMArena top-of-board moved less than 1 Elo.
Issue 4 · 2026-05-17
Frontier text took a breath. The constraint stack moved to supply, regulation, and pricing structure.
The model layer took a breath this week.
Issue 4 · 2026-05-16
Subagents turned delegation from a single conversation into an agent team pattern.
W20's operating pattern is role separation.
Issue 3 · 2026-05-09
Agentic work became a review problem: the scarce skill is defining good and done.
W19's technique read is verification-first delegation.
Issue 3 · 2026-05-09
Agentic applications became capacity products: more compute turned directly into more work.
W19's application-layer signal was that agent products are now constrained by throughput, not just model quality.
Issue 3 · 2026-05-08
Frontier text went quiet. Voice crossed into chain-of-thought. The week's biggest capability move was a compute deal.
W19 was the week the model layer split.
Issue 3 · 2026-05-08
Capability stops being about parameters. The constraint stack flipped to power, fiber, and packaging.
W19 is the week the AI compute story stopped being told in parameter counts.
Issue 2 · 2026-05-02
The agent operating layer moved from prompt craft to repeatable delegation.
The first read starts where the market was already shifting: the interesting unit was no longer the best single prompt, but the repeatable way a human delegates work, supplies context, checks the answer, and teaches the system what to re...
Issue 2 · 2026-05-02
AI-native SaaS stopped being an assistant story and became a workflow packaging story.
W18 brought the first backfill week where application-layer evidence was explicit.
Issue 2 · 2026-05-01
Six new tree rows, two distribution earthquakes, and the open-weights ceiling closing to 6 points behind GPT-5.5.
W18 is the model layer's first true cadence test under the new weekly Pulse: six new tree rows across five vendors, two distribution earthquakes (Microsoft / OpenAI exclusivity formally ends, GPT-5.5 + Codex + Bedrock Managed Agents land...
Issue 2 · 2026-05-01
Q1 broke the bear thesis. The marginal dollar started earning more — and the network still compounds.
The W17 read was 'the marginal dollar earns less at every layer except the network.' Q1 2026 hyperscaler prints (MSFT / GOOG / META / AMZN, all on Apr 29) reset the math.
Issue 1 · 2026-04-25
Agentic work crossed from prompt craft into capability-gated operating practice.
The first operating-layer read starts with the week agentic systems became a governance object.
Issue 1 · 2026-04-25
The application layer's first pressure point was not UI replacement; it was governed action.
W17's application-layer read starts upstream of the later SaaS earnings wave.
Issue 1 · 2026-04-25
April rewrote the floor: open weights crossed the closed frontier, and the safety-gated frontier became a procurement criterion.
April 2026 was the most productive month in frontier model history.
Issue 1 · 2026-04-25
The marginal dollar at every layer earns less. Except the network.
Capital is still flooding the AI stack, but the marginal dollar earns less than the prior dollar at every layer except one.
Market map
AI Market Reference Architecture
A NIST-cloud-style taxonomy for AI market boundaries, control points, shared responsibility, and leaders.
Scenario engine
AI Future Studio
Set a horizon, tune the forces, and generate a grounded briefing on where the AI market could bend next.
Inflection history
AI Shockwave Timeline
The events that reset frontier assumptions, with before/after impact deltas across the AI landscape.
Working framework
The Thesis
How we read the AI compute stack: three lenses, four laws, one flywheel. The framework filters signal from noise and names what would change our minds.
Model lineage
LLM Evolutionary Tree
Track frontier and open-weight model families as capabilities branch, converge, and reshape the stack.