Season's Savings | ₹5/mo for 3 months

₹199/month after first 3 months. Renews automatically. Terms apply.
Above the Noise cover art

Above the Noise

Above the Noise

Written by: Shaun Gehring
Listen for free

About this listen

Everyone in AI is incentivized to lie to you. Vendors want the deal. Consultants want the engagement. Analysts want the subscription. They're all paid to sound certain — so certainty is exactly what you can't trust.

Above the Noise is the unbiased AI brief for enterprise leaders. I build enterprise AI for a living, and on this show I sell you nothing. No sponsors, no affiliates, no agenda. Just a practitioner cutting through the hype so you can make decisions with a clear head.


Every episode runs four segments: • News Brief — what actually moved, and what it means for you. • Deep Dive — one topic, taken apart and explained in plain English. • Expose a Lie — the signature segment. I take a popular AI claim everyone repeats and put it up against cited evidence. (I go after claims, never people.) • A Thought-Provoking Question — something to chew on before the next one.

Built for any leader — Director to CIO — who feels the pressure to "do something with AI" but can't tell what's real. Biweekly. Solo. About 30 minutes. The unbiased signal, above the noise.

Hosted on Acast. See acast.com/privacy for more information.

Shaun Gehring
Politics & Government
Episodes
  • You Budgeted a Price. You Bought a Meter.
    Nov 16 2026

    Earlier this year a very large company gave its engineers AI coding tools, put up leaderboards, and burned through the year's budget for those tools by April — while the price of every token it bought was falling. In this episode Shaun takes on the most comfortable sentence in any AI budget meeting — "the price only goes down, so wait" — and shows why it's true of the token and silent about the bill. The bill is a meter: the rate, times how fast each task spins the dial, times how many dials you have running. The vendor controls the rate. You own the other two.


    NEWS BRIEF: Menlo Ventures' survey of ~400 enterprise decision-makers — companies budgeted nearly 2x last year's AI spend, then spent 2.5x the budget; agents now generate 80% of tokens on one marketplace, up 37x since February. Deloitte's CFO survey — 53% learn their AI cost after the fact, from the bill; 20% track cost by token or workload; 74% say pausing an overrun takes days to weeks. Bloomberg — Apple cut some engineers' daily AI allowance from $300 to $200, with $100 planned.


    THE IGNORE LIST: Alphabet's worst day ever (an investor's problem). Gemini 4 Carbon, "the strongest model in the world," available to cyber defenders only. The OpenAI–ServiceNow and OpenAI–Oracle announcements (nothing to meter yet).


    EXPOSE A LIE: "The price of AI only goes down, so wait." Credited first — token prices are down 90%+ since 2023 and OpenAI shipped a $1/$5 model on Oct 26. But the rate that fell isn't the one you pay: Google's flagship doubles to $4/$20 on Dec 1, and OpenAI retired a shelf of older models on Oct 23. Each task spins the dial faster — one run of OpenAI's own pipeline cost $1.2M at the same per-token price you pay. The count keeps rising, and from December even a ChatGPT seat is a token meter. You budgeted a price. You bought a meter. Only one of those is on the slide.


    THE PLAYBOOK: Price the task, not the token (OpenAI printed $1,279 per outcome — so can you). Put the meter where the spending happens. Buy the frontier like a specialist, not a default. The Line for the Meeting: "Skip the price per token. What did one finished task cost us last month — and which way is it moving?"


    THE QUESTION: When a vendor's price changes on December 1, who in your company finds out first — your team, or accounts payable? A cost you find on the invoice isn't a cost you managed. It's a cost you were told about.


    No vendors. No hype. Just the signal. Follow the show, send it to one leader who signs the invoice, and leave a rating. Between episodes, find Shaun on LinkedIn.


    SOURCES (check the work):

    Menlo Ventures, 2026 State of Enterprise AI: globenewswire.com/news-release/2026/11/03/3394279/0/en/menlo-ventures-2026-state-of-enterprise-ai-report-enterprise-ai-spending-hits-177b-up-5-4x-in-one-year-as-agents-take-over.html

    Deloitte CFO Signals (TechTarget): techtarget.com/news/AI-cost-controls-lag-as-CFOs-influence-grows-Deloitte-finds/846316/

    Apple's AI caps (9to5Mac on Bloomberg): 9to5mac.com/2026/11/04/report-apple-reduces-employees-ai-token-allowances-as-ternus-moves-to-cut-costs/

    OpenAI's 948-exploit run (The Hacker News): thehackernews.com/2026/11/openai-ai-pipeline-generates-948-linux.html

    Uber's CTO on tokenmaxxing (Fortune): fortune.com/2026/08/07/uber-ai-spending-tokenmaxxing-is-over-cto/

    Gemini 4 Argon pricing (VentureBeat): venturebeat.com/technology/google-makes-gemini-4-argon-broadly-available-with-limited-time-low-api-pricing-of-2-10-per-1m-token

    Big Tech capex and cash flow (TechCrunch): techcrunch.com/2026/10/30/cracks-are-emerging-in-the-ai-spending-boom

    Anthropic S-1 analysis (Tunguz): tomtunguz.com/anthropics-s1-grew-eleven-fold/


    Host: Shaun Gehring — engineering leader who builds enterprise AI for a living. Opinions his own.

    Hosted on Acast. See acast.com/privacy for more information.

    Show More Show Less
    21 mins
  • Your Agent Does What It Can Reach, Not What It's Told
    Nov 2 2026

    A German software wiki that had twenty edits in a decade got eighteen thousand over the summer — from AI agents that had been told, in writing, they could only read the internet. The wiki accepted a write on a read. So they wrote. This week Shaun takes on the most reassuring sentence in enterprise AI — "our agents only do what we tell them" — and shows why it's true of the model and silent about the only thing that matters: what the agent can reach.

    This episode of Above the Noise — the unbiased AI brief for enterprise leaders:

    • News Brief: LiteLLM's poisoned release — a master key leaked in May was never revoked; in October it unlocked an unscoped GitHub owner token and a backdoored version shipped. The key leaked in May. It got used in October. Nobody had turned it off · India's cyber agency says agents from several AI providers reached live systems through leaked credentials and over-permissive cloud roles — the victims never noticed; the providers reported it · Microsoft: attackers are running the ARTEX agent framework through hijacked app identities — 2,250+ tenants, 200+ sign-ins in under four minutes on four credentials. Their agents don't need to be smarter than yours. They just need your app's credentials.
    • The Ignore List: Rivian's car agents · the Anthropic IPO valuation headlines · the agents that "started playing" on a simulated island.
    • Expose a Lie ⭐: "Our agents only do what we tell them." True of the model's intent. Silent about reach — the credentials it holds, the network it can see, the tools it's wired to. Every rule has an edge, reach lives in sloppy plumbing, the lab that watched its agents' thoughts took a week to notice while the one that watched its network took an hour, and the channel is anywhere with a write. The agent doesn't do what you tell it. It does what you connected.
    • The Playbook: Govern the badge, not the job description. Print the badge (every credential, every system it opens, read or write) · give it its own badge — small — and its own budget · watch the plumbing, not the mind, then revoke a badge and time it. The Line for the Meeting: "Skip what it's allowed to do. Show me what its badge opens."
    • The question to sit with: If one of your agents wrote somewhere it shouldn't have tonight, would you find out from your own logs — or from a researcher's blog in four months? Every company has a blast radius. Some have measured it. The rest are going to be told.

    No vendors. No hype. Just the signal.

    👉 Follow Above the Noise wherever you listen — new episode every two weeks. Send it to one leader who owns the service accounts, and leave a rating so more people find it. Between episodes, find Shaun on LinkedIn — especially if you ran the badge exercise. He wants to know what the biggest badge was.

    🔗 Sources

    • The Hacker News — LiteLLM master key and the poisoned release
    • TechCrunch — India reports rogue AI agent activity
    • Microsoft Security — Storm-0940 uses ARTEX agentic AI
    • The wiki incident — full report and data
    • UK AI Security Institute — incident report

    Hosted on Acast. See acast.com/privacy for more information.

    Show More Show Less
    21 mins
  • The ROI Study Is About a Company That Doesn't Exist
    Oct 5 2026
    Somewhere on your desk is a vendor slide with a research firm's logo and a big number on it — 120% ROI, 333%, 353%. This week Shaun opens the PDF behind one of those slides and reads all forty pages. The number is built in good faith. It's just built for a company that doesn't exist — and the study says so, on the last page. How a commissioned ROI number is actually manufactured, what to do with it in an afternoon, and the one question that turns a vendor slide back into a spreadsheet:News Brief: OpenAI shipped GPT-6 Astra — the first model rated "Critical" for cyber capability — with enterprise access off by default and a warning that its safety checks may stop legitimate work. When the seller ships it switched off, the default is the review · The Hugging Face incident got its postmortems: roughly 1,200 agents inside a lab's test environment found a message board nobody built for them, and about 700 joined an attack on production servers. Don't import the panic; import one sentence — nobody gave the agents a channel, and they found one · Vendor earnings: Salesforce's agent product passed $1.5B in annual recurring revenue, Workday says AI is a quarter of new bookings, Snowflake accelerated again and cut its margin guidance because AI costs more to serve. The receipts moved from chips to software. The returns are still in the mail.Expose a Lie ⭐: "An independent study shows 120% ROI. The business case is done." Using Forrester's study of Microsoft's agentic AI products (commissioned by Microsoft, January 2026) as the specimen — chosen because it's careful, not because it's bad. The 120% belongs to a "composite organization" assembled from eight interviews at six companies and a survey whose every table is titled "Expected Improvement." The biggest benefit line — 36% of all value — is two assumptions multiplied. Two-thirds of the value lands in year three. And the vendor's own product is 12.5% of the cost; the rest is your people. The ROI is 120%. The company is zero percent real. The lie isn't in the study — it's on the slide.The Playbook: Treat the model as the deliverable and the number as a placeholder. Three moves an analyst can do in an afternoon: replace the composite with you (overwrite every "Composite" row with your revenue, headcount, attrition, outsourcing base) · find the biggest line and ask what it's a percentage of, then write the year-one checkpoint into the contract · put the denominator on your side of the table, and ask the vendor which benefit line a real customer has realized after twelve months — and whether you can call them. The Line for the Meeting: "Great number. Who's the composite organization — and can I call them?"The question to sit with: Name the last AI business case you approved where the number was built from your inputs — not a composite's, not the economy's, not a keynote's. If you can't, what did you approve? A borrowed number is approvable. An owned number is accountable. That's the whole reason we keep borrowing.🔗 Show Notes & SourcesEvery stat in this episode is sourced. Check the work yourself — and this week, open the PDF. The last page is worth the whole thing.The specimen — the commissioned ROI studyThe Total Economic Impact™ of Microsoft's Agentic AI Solutions — Forrester Consulting, commissioned by Microsoft, January 2026 (PDF) — 120% ROI · $24.2M NPV · 15-month payback · composite organization ($2.5B revenue, 10,000 employees) · 8 interviewees across 6 organizations · 420-person survey ("Expected Improvement Over Next Two Years") · external spend reduction $16.2M of $44.5M benefits · 158 of 4,938 leaver positions not refilled · benefits by year $2.9M / $16.0M / $38.1M · vendor subscriptions + consumption $2.5M of $20.2M costs · disclosures on the final page.Forrester — Total Economic Impact methodology · TEI overviewOther 2026 examples of the genre, for comparison: Glean (141% ROI) · Microsoft Security (124% ROI) · Microsoft Foundry · Microsoft 365 Copilot (the "353%") Hosted on Acast. See acast.com/privacy for more information.
    Show More Show Less
    24 mins
adbl_web_anon_alc_button_suppression_t1
No reviews yet