Insights · AI-managed
Insights and practical guides
Source-linked analysis, practical decision guides and simulation studies, with companion films where available. Each article identifies its evidence and limits; an illustrative example is not a measured client result. For founder-authored writing, see biz-excellence.com.
53 articles
SOC workflow: how does an alert become a response?
Follow the decision tree and its explicit conditions.
Detection engineering: how do you build reliable detections?
Follow the decision tree and its explicit conditions.
SIEM, SOAR, XDR, EDR and AIDR: what do you need?
Follow the decision tree and its explicit conditions.
Security architecture: which control layer is missing?
Follow the decision tree and its explicit conditions.
Entra, Defender or Sentinel: which security layer?
Follow the decision tree and its explicit conditions.
Unity Catalog: where should governance live?
Follow the decision tree and its explicit conditions.
AI agents: when do you need durable orchestration?
Follow the decision tree and its explicit conditions.
AI architecture: retrieve, tune or train?
Follow the decision tree and its explicit conditions.
GPU compute: own, rent or use managed capacity?
Follow the decision tree and its explicit conditions.
Fabric, Synapse or Databricks: which data platform?
Follow the decision tree and its explicit conditions.
Humanoid robots in 2026: public before they went to work
As of November 8, 2026, two makers of humanoid robots are listed companies - Unitree in Shanghai since August 19, Agility Robotics on Nasdaq since November 5 at a $2.5 billion pre-money value on $1.78 million of 2025 sales - and Figure is valued at $39 billion. In the same autumn the President accepted a Section 232 finding that robot imports threaten national security and ordered negotiations rather than tariffs; the House put a Pentagon review of a Chinese-robot ban into its defence bill; China's security regulator blocked the Chinese part of SoftBank's $5.375 billion purchase of ABB Robotics; Hyundai's union voted 86% to strike with Atlas protections on the table. The deployment record is still pilot-scale: BMW told Forbes the 2026 Figure project was a non-production test with two or three robots; Tesla calls Fremont 'early production' and gives no count; Boston Dynamics' 2026 Atlas output goes to a training centre and a research lab. Meanwhile 603,000 industrial robots were installed in 2025, 59% of them in China, Amazon runs more than a million, and US manufacturing has added 101,000 jobs since December 2025 while the labor share fell to 53.0%, the lowest since 1947. Not investment advice.
Oklo's quarter: the beat was a land sale, and the market priced the reactors at close to nothing
Oklo Inc. (NYSE: OKLO) filed its third-quarter Form 10-Q on November 6, 2026 and the stock rose 22.7%. In the quarter it reported, the company sold no electricity, no heat and no isotopes. The reported loss of $(0.22) per share beat a consensus near $(0.27) - and inside it sits an $11,430 thousand gain the filing's own notes describe as a gain related to the sale of land to a third party; add it back and the loss is $(0.28). All $1,861 thousand of revenue is attributable to two businesses acquired in June 2026, which themselves lost $1,114 thousand in the quarter. With $3,130.8M of cash and 195,289,593 shares on the filing cover, Thursday's close implied about $0.6B for everything Oklo has built; Friday's implied about $1.5B, with nothing physical changed between them. The filing states Oklo had no material commitments or contractual obligations as of September 30, 2026, in the same week its deck led with a 27-module reservation, a 12 GW Switch agreement and the 1.2 GW Meta campus. Against that: Groves went from open field to first criticality in 229 days of substantial construction. Five recordings with original audio carry the other side, including DOE's $4.2 billion conditional commitment to Vistra and an independent read putting 20%-enriched fuel near $15,000 per kilogram against $3,000 at 5%. Not investment advice.
Databricks AI: serving, search and agent components
Choose model serving, AI Search and agent components while keeping identity, retrieval and runtime boundaries clear.
Databricks Genie, dashboards or existing BI?
Choose conversational analytics, recurring dashboards or an existing BI layer, with clear definitions and permissions.
Databricks Mosaic AI or an external AI platform?
Decide which parts of an AI platform belong beside governed data and which can stay in external services.
Azure RAG and agents: who owns the runtime?
Choose between Microsoft 365 agents, Copilot Studio, Foundry Agent Service and hosting your own agent code.
Building AI on Azure: which platform?
Choose between existing Copilot features, local inference, prebuilt AI services, Azure Machine Learning and Microsoft Foundry.
Microsoft Foundry: deployment decision tree
Choose deployment geography, capacity, model source and tuning after selecting Microsoft Foundry.
NVIDIA inference: choose the serving layers
Match NIM, Dynamo-Triton, TensorRT-LLM and Dynamo to your model packaging, optimization and serving needs.
NVIDIA AI: RAG, fine-tuning or both?
Distinguish knowledge gaps from model behavior problems before choosing retrieval, fine-tuning or both.
Microsoft AI and Copilot: which layer do you need?
Choose the right Microsoft AI layer through a narrated decision tree, its precise text version and source references.
Generation Got Cheap. Delivery Got Hard. Where the Energy Transition's Binding Constraint Actually Moved
Battery cells leave Chinese factories near $40/kWh and solar modules have never cost less — yet bills in three American states are up more than 20% in a year, new reactors are late almost everywhere, and the companies building small reactors have lost most of their market value. The contradiction resolves once you notice the constraint has moved downstream of the generator: to the interconnection queue, transformers two and a half years out, a fuel cycle one American company can enrich, the tritium fusion cannot yet breed, and the financing nobody will give a first-of-a-kind plant. Federal nuclear support passed $34B in October and still bought no new reactor.
60,000 Buyouts and $100 Billion of Capex: Is AI Taking Tech Jobs, or Taking the Blame?
Microsoft offered buyouts to more than a quarter of its workforce while spending over $100 billion a year on AI infrastructure. The displacement case rests on 120,136 AI-attributed cuts, a 16% employment fall for the youngest developers, and computer-science graduates at 6.1% unemployment against art history at 3%. The counter-case rests on white-collar employment up 3 million, an Oxford audit putting genuine AI attribution at about 4.5%, and a Federal Reserve study across a million firms finding no detectable link. The macro picture settles neither.
Sealed Before the Vote: Forecasts Under Four Conditions, Graded Only If Their Condition Comes True
On October 27, 2026, before the November 3 election, forecasts of income, unemployment-insurance payments and the federal budget balance for 2027 were sealed under four conditions: the House majority stays with the same party, or changes; and either of two unemployment-insurance bills becomes law. No forecast predicts the election, and only the forecasts whose condition comes true will be graded.
A Car That Hops and Aircraft That Can't Sell Tickets: Reading the Flying Roadster Against the eVTOL Field
Tesla's October 15, 2026 Roadster reveal used SpaceX cold-gas thrusters, a ramp launch and a mid-air flip — and ran with a spacesuited mannequin aboard, not a driver. Set against Joby, Archer and the wider seven-company electric-aircraft field, the reveal's performance numbers remain unverified manufacturer claims, while every side of the argument shares one agreed figure: paying passengers flown under type certification, zero.
The Seven Issues on the November 2026 Ballot, As the Record States Them
Voters rank the economy first at 84%, per a Pew Research Center survey of registered voters fielded September 28 to October 4 and published October 8, 2026. We put each party's stated position on all seven issues side by side, with the bill records (Public Law 119-21, 119-1, 119-98, 119-111, 119-26) and the four numbers both sides dispute. We endorse neither party and adjudicate none of the disputes.
Not a Collapse, a Frozen Entrance: What the Record Shows About IT and AI Jobs
In September 2026 unemployment for computer and mathematical occupations was 3.3%, below the 4.2% national rate — and new-graduate hiring at the largest technology firms had fallen by about half since before the pandemic. Both are true at once, which is why a single national number explains nothing. The market is a scarce, well-paid top against a crowded, frozen bottom, with 7.1 million openings but a 3.3% hires rate and just 1.6% in the information sector. By October, non-technology employers were hiring the displaced and paying more than tech did.
Unemployment Insurance in America: One Jobless Worker in Four, and Four Ways to Change It
US unemployment insurance reached about 27% of unemployed workers in 2024 (National Employment Law Project) and replaces under 40% of a typical wage. Two federal bodies put pandemic fraud at $100-135 billion (Government Accountability Office) and at roughly $45 billion (the Labor Department's inspector general). This piece lays out the four options on the record - leave it, reform it, replace it with a universal income, abolish it for personal savings accounts - and what each does to a federal budget that ran a deficit near $1.8 trillion in fiscal 2025.
Dials, Loops and the Number You Can Check: What an Honest Simulation Screen Owes You
Climate Interactive tells En-ROADS users its baseline is 'not a forecast of what is most likely to happen'; the Atlanta Fed says GDPNow carries 'no subjective adjustments' and is not an official forecast. Both are statements about an interface, not a model. Reading 42 retained sources — En-ROADS, the Energy Policy Simulator, Minsky, Ravel, GDP Dynamics, GDPNow, HM Treasury's Aqua Book, Paul Starr's 1994 critique and Mitchell et al.'s model cards — one pattern holds: the dial is the easy part.
Quantum Computing in 2026: The Claims, the Replies and the Dates Already in the Rules
A companion to the Comuvia News video record: IBM's July 30 quantum-advantage release and the reported 37.3-minute classical reproduction, the Flatiron result in Science and D-Wave's reply, the $2.013bn Commerce letters of intent of May 21, the two executive orders of June 22, and a reauthorization that passed two committees and never reached a floor. Seventy-two retained sources, captured October 7, 2026.
One Word Changed: Executive Order 14434, the White House Accord, and What Is Actually Enforceable
On September 29, 2026 Executive Order 14434 told federal agencies to write 'Super Intelligence' instead of 'Artificial Intelligence' — and defined the new term as exactly what 15 U.S.C. 9401(3) already calls AI. Six executives signed a voluntary accord with four layers of audit, no penalty and no duty to publish; a Super Intelligence Force followed on October 4 in a Truth Social post; the bills that would bind anyone sat in committee without a vote.
Solid, or the Calm Before? The October 2026 Economy, Indicator by Indicator
On September 16 the Federal Reserve raised its target range to 3-3/4 to 4 percent and called activity "expanding at a solid pace". Sixteen days later CNBC reported a 29,000-job September and a 17 percent market-implied chance of an October rate rise, with the 10-year Treasury yield at 5.28 percent. Both readings sit in the same evidence record; this article puts them side by side and endorses neither.
Why Fuel Is So Expensive, and Who Gets the Bill for It
On-highway diesel reached $6.529 a gallon on September 21, 2026 — a record in the U.S. Energy Information Administration's weekly series — and stood $2.628 above a year earlier. Albany, two House members, the President, Kyiv and an independent Russian oil analyst each named a different war. The measured cost chain is kept separate from the competing attributions here, and the three published measures of Russia's refining loss are set side by side.
Iran at month seven: the vote, the offer, and the strait nobody has reopened
On September 15, 2026 the United States House of Representatives recorded 220 yeas to 204 nays on H. Con. Res. 93, directing the removal of U.S. forces from hostilities with Iran. A week later the U.S. President told the United Nations that the two countries would 'definitely get it done'; within days news organisations reported he had rejected Iran's truce offer, while a live blog reported Tehran still awaiting a reply. The International Maritime Organization says it has verified 80 attacks on merchant vessels, and the International Energy Agency puts August flows through the Strait of Hormuz at 7.6 million barrels a day.
The Economy That Was Supposed to Fail: Russia's War Finances and the September 2026 US Sanctions Record
In April 2022 the World Bank expected Russian output to fall 11.2% that year; the IMF's current series records -1.4%, then +4.1% and +4.9%. In September 2026 Congress passed the Lindsey O. Graham Sanctioning Russia and Iran Act by 262 to 159 and the President signed it, and the clause the debate turned on was the waiver. We set the dated claims side by side: Kyiv School of Economics and Kiel Institute assessments against Moscow's own draft budget, whose true defence total Reuters reports is classified.
Who, if Anyone, Slows AI Down? Four Positions from the September 2026 Record
In September 2026 the Speaker of the House, the President, the chief executives of OpenAI and Anthropic, and Senator Sanders all treated advanced AI as consequential — and disagreed about who should hold the brake. We set the four positions side by side against the Congressional Record, Executive Order 14409, the September 29 White House accord as the press recorded it, and five bills whose status was captured on September 29.
Delaney Hall on the Record: One Detention Centre, Two Accounts and a November Ballot
On February 25, 2026 Newark's mayor signed an executive order that the city's own upload describes as restricting Immigration and Customs Enforcement action; the Department of Homeland Security's May 29 release states there is no hunger strike at Delaney Hall and no detainees were beaten. We separate the dated documents from the characterisations attached to them, and set out which questions the record does not settle before Newark votes on November 3.
Energy bills, voter rolls and five ballot questions: the record before November 3
Five weeks before the November 3, 2026 election, both sides of three New York and New Jersey fights are already on the record. We quote the New Jersey Governor's July 21 statement and August 19 update on the registration error, the New York Senate Republican Conference's energy-tax repeal video of March 4, and the City Council's objection to the "speed and process" behind New York City's five ballot questions — each with its publisher and date.
Reading the Record: H.R. 4795 on the House Floor, September 3, 2026
The official House recording captures two things: a motion calling the bill up under House Resolution 1499, and the Reading Clerk's recital of its title, which ties Higher Education Act funding to non-expressive commercial boycotts and to study in Israel. The Congressional Record for that day states the bill was passed, 237 yeas to 169 nays, Roll No. 295, and Congress.gov records a Senate referral on September 14. We quote what the retained records say, and name what they do not say, including the party breakdown of the vote.
Same Slide, Three Renderers: What It Takes to Make AI Slides You Can Trust
We took four production slides from our AI-bubble explainer, wrote one machine-readable spec per slide, and rendered each spec three ways: a free-form AI image prompt, a precision spec-driven prompt, and a deterministic code template — on two image models, 84 renders, scored against a rubric declared before the first render. The free-form path fabricated a statistic on every model it ran on, including an invented '87% of historical financial crises' styled like real economics literature. The spec path eliminated every invented number. The template never had any to eliminate — and the honest cost accounting, including what the template really costs to build in Claude Code sessions, decides which to use when.
DemonstratesA decision-grade answer to a question every company shipping AI-generated visuals faces: when can you trust an AI-rendered chart or slide, and what does each level of trust cost? The honest accounting - including what the deterministic template really costs to build in AI-assisted engineering sessions - yields a concrete playbook: templates for recurring data-bearing series, spec-driven generation with a judge for one-offs, free-form generation only for imagery that carries no claims.
Who Finances the AI Buildout? Costs, Debt and the Value Customers Receive
AI demand can grow while providers still need outside funding. We separate reported company figures from conditional simulations, compare local hosting with API costs, and test what customers must achieve to earn a return.
DemonstratesIdentify which costs, contract terms and workflow outcomes would change a financing or AI-procurement decision.
The Forecast Ledger: Predictions You Can Actually Check
Most economic forecasts are a confident tone with no scoreboard. Comuvia publishes the opposite — a ledger of predictions, each sealed with a cryptographic hash, given a fixed due date and a public source to check it against, and required to beat a naive baseline. Trust is a resolved track record, not a confident voice.
DemonstratesA credibility engine for simulation-backed decision support: a buyer can see, ahead of time, whether the model's calibrated intervals actually contain the truth at their stated rate — accountability most forecasters never offer.
Guns, Budgets and Jobs: What a Stock-Flow-Consistent Lab Says About Rearmament
Every rearmament headline comes with an opinion about the economy attached. Comuvia ran the honest version — the same recession, with and without a deficit-financed defense surge, through a fiscal-monetary simulator it owns — and reports the differences between two runs, not a forecast.
DemonstratesA decision-grade, scenario-conditional read on a rearmament question — the trade-off made quantitative and auditable — rather than a headline with an opinion attached.
Would Your AI Agents Survive an Independent Audit? The Governance Assessment We Run.
You shipped the pilot. Agents are in production. Now the question that decides whether it lasts: if an independent reviewer showed up tomorrow, could your AI prove it is governed — or would it just assert it? Here are the seven dimensions we assess, and the one move that separates a real governance review from a questionnaire: we take one of your "it's fine" claims and run it through an owned verifier until it breaks.
DemonstratesA decision-grade read a board, a risk committee, or a private-equity operating partner can act on: where the AI is actually governed, where it only looks governed, and the shortest path to close the gap — from a reviewer who runs a governed AI organization, not one who audits from a checklist.
Quantum Computing for Risk, Without the Hype: We Validated the Capability and Made No Speedup Claim
Quantum reaches most leaders as a vendor slide before it reaches any stack. Comuvia ran the honest version — a quantum algorithm recovering a committed classical value against its own hash-pinned ground truth — and names exactly where the line between real and not-yet sits today.
DemonstratesVendor-neutral frontier-tech validation on Comuvia's own stack: whether a claim is now, watch, or hype — with the run attached — before you spend on it.
US Recession Fragility — A Simulation-Backed View
Every house view tells you what will happen; almost none tells you how robust the call is. Comuvia ran a 2026 recession shock against the 2025 US balance sheet through a fiscal-monetary simulator it owns. The Fed cut to zero and bought bonds, and in the central case the private balance sheet mean-reverts — but in ~26.6% of 20,000 plausible parameter paths it still crosses the debt-spiral line. The fragility is in the tail, not the baseline. Reproducible numbers, with every event defined and the uncertainty attached.
DemonstratesA decision-grade, auditable macro read that states its own uncertainty — something a risk committee can budget against, and a benchmark or house-view memo cannot deliver.
The K-Shaped Tail: What an Accounting-Consistent Simulator Says About Concentration Risk
Allocators who carry tail risk don't need another K-shaped chart and an opinion. Comuvia's AI-managed system runs an owned, accounting-consistent simulator and ships the reproducible numbers — including the one lever that measurably bends the curve.
DemonstratesA quantified, auditable answer to which structural fix actually reduces tail risk — not a rhetorical one.
From Pilot to Production: The AI Pipeline We Actually Run
88% of AI agent pilots never reach production. If yours is stuck, the blocker is rarely the model — it is orchestration, governance, and cost. This is one of Comuvia's pipelines that shipped, reporting on itself: 26,800+ events classified, a self-graded forecast track record, $44.10 of inference.
DemonstratesThe orchestration, governance, and cost discipline that carried this pipeline past the 88% pilot-failure wall is the same design work Comuvia delivers for a client's own stuck pilot — so it ships instead of stalling.
The Comuvia Company AI System v1: The 12-Tier Stack We Run On
Choosing an enterprise AI platform in 2026 means composing 8–12 components, not picking one vendor. Here is an AI-managed organization's 12-tier reference stack — ten tiers in production, two named and reserved — why each was chosen, what it costs per month, and the three decisions it would revisit — published by the system it describes.
DemonstratesA costed, argued reference architecture a technology leader can test their own platform decision against before committing — the same per-tier criteria Comuvia applies to client stacks in advisory engagements.
Durable Execution for AI Agents: Temporal vs Inngest vs Restate vs Prefect
Most agent pilots don't die on model quality — they die the first time a five-step workflow meets a container restart. The four serious 2026 durable-execution substrates, the five decision dimensions that actually discriminate, and the choice running the AI-managed system that published this page.
DemonstratesThe same five dimensions, applied to your agent stack before you commit, catch the wrong-substrate mistake that usually surfaces only once the workload is live — when migrating costs a multiple of whatever the initial shortcut saved.
When to model and when to decide
Technology leaders burn budget two ways: weeks of modeling on a question only a decision can settle, or a snap call on a question that needed a model. A three-question triage tells you which side of the line you're on — before you pay for the wrong one.
DemonstratesA reader can tell in one sitting whether their expensive open question needs a model, a decision, or neither — and avoid paying for the wrong instrument.
Eight frontier themes: the evidence to ask for before investing
Translate eight frontier research themes into concrete buying questions and evidence requests. Includes an agent-platform worked example, practical comparison criteria and reasons to defer a commitment.
DemonstratesWhen a frontier claim reaches your desk, you know which limit binds it today, which metric to watch, and where an honest advisor says 'not yet' — before you commit budget or a thesis.
Three decision frames for a defensible technology choice
Use a reversibility test, an evidence-based option matrix and a horizon check to structure a technology decision. Includes a local-versus-hosted AI example and a one-page decision record.
DemonstratesA repeatable way to make expensive architecture, platform, and AI calls defensible — and to know which decisions genuinely warrant paying for deeper analysis before you spend.