News Weekly
LV 10 XP
0% read
Home › Issues › Issue #1 › S38
S38enterprise
#38 Issue #1Confirmed

Jensen Huang's Dreamforce remarks: agents as the new category of enterprise software

On 2026-09-15, NVIDIA CEO Jensen Huang joined Salesforce CEO Marc Benioff on the Dreamforce keynote stage at Moscone Center, San Francisco. CNBC reported roughly 12,000 people attended the event; OpenAI CEO Sam Altman also appeared at Dreamforce, and Anthropic CEO Dario Amodei shared the stage around the same event, where he and Huang publicly diverged on AI safety (CNBC). Salesforce and NVIDIA announced Koa, described by Salesforce as its first CRM reasoning model for Agentforce, built on NVIDIA Nemotron (Salesforce press release; corroborated by TechCrunch). Huang argued AI has become "an infrastructure layer across the planet," that "every company, every enterprise, every country would become an AI company," and "It's going to be an agentic enterprise." He closed with "Engage AI. Don't get left behind." (NVIDIA blog, 2026-09-15). Huang rejected the "SaaSpocalypse" thesis: "Don't give up on the technology... I was the first CEO to come out not in the software industry to say the end of software is nonsense." He and Benioff agreed AI "is going to be a layer on top of software" that is "agentic" and "going to make the software so much better" (TechRadar, 2026-09-15). On safety and regulation Huang said "Safety is an engineering problem, not a legal one," and "The market forces are already there. We don't need any new laws. We don't need new regulations." (TechCrunch, 2026-09-15). The NVIDIA blog similarly recorded "Safety is paramount in a lot of ways. It's job one... However, safety is an engineering problem..." QZ (2026-09-16) characterizes Huang as rejecting AI-slowdown calls at Dreamforce. FT headline (2026-09-16, paywalled): "Nvidia and Meta bosses reject efforts to co-ordinate AI slowdown."

NNVIDIA·Sep 15, 2026·39 min read
A thin luminous layer settles over a vast city of software blocks like a second skin, adding structures without obscuring the buildings beneath.
How do you want to read this?

Tailored emphasis while keeping the full article available.

🎓Understand it
ExplorerThe story in plain words⌘Build with it
BuilderHow it works, how to try it▥Make decisions
Decision makerCosts, risks, next moves
Best for you · Explorer

🎓 Start with the story, why it matters, and where it goes next.

🎓What happened?↓🎓Before → Change → After↓🎓Why it matters↓🎓What became possible?↓
At a glance

The essential information in 30 seconds

What happened
  • FACT: On 2026-09-15, NVIDIA CEO Jensen Huang joined Salesforce CEO Marc Benioff on the Dreamforce keynote stage at Moscone Center, San Francisco. CNBC reported roughly 12,000 people attended the event; OpenAI CEO Sam Altman also appeared at Dreamforce, and Anthropic CEO Dario Amodei shared the stage around the same event, where he and Huang publicly diverged on AI safety (CNBC).
  • FACT: Salesforce and NVIDIA announced Koa, described by Salesforce as its first CRM reasoning model for Agentforce, built on NVIDIA Nemotron (Salesforce press release; corroborated by TechCrunch).
  • COMPANY CLAIM (attributed): Huang argued AI has become "an infrastructure layer across the planet," that "every company, every enterprise, every country would become an AI company," and "It's going to be an agentic enterprise." He closed with "Engage AI. Don't get left behind." (NVIDIA blog, 2026-09-15).
  • COMPANY CLAIM (attributed): Huang rejected the "SaaSpocalypse" thesis: "Don't give up on the technology... I was the first CEO to come out not in the software industry to say the end of software is nonsense." He and Benioff agreed AI "is going to be a layer on top of software" that is "agentic" and "going to make the software so much better" (TechRadar, 2026-09-15).
  • COMPANY CLAIM (attributed): On safety and regulation Huang said "Safety is an engineering problem, not a legal one," and "The market forces are already there. We don't need any new laws. We don't need new regulations." (TechCrunch, 2026-09-15). The NVIDIA blog similarly recorded "Safety is paramount in a lot of ways. It's job one... However, safety is an engineering problem..."
  • INDEPENDENT EVIDENCE: QZ (2026-09-16) characterizes Huang as rejecting AI-slowdown calls at Dreamforce. FT headline (2026-09-16, paywalled): "Nvidia and Meta bosses reject efforts to co-ordinate AI slowdown."
Why it matters
  • Enterprise impact is the highest dimension (discovery: enterprise_impact 8): CRM is the front office of most companies; a reasoning model that can act (update opportunities, route cases, schedule follow-ups) inside Agentforce changes what "CRM software" does.
  • It reframes the SaaSpocalypse debate with CEO-level authority on both sides (Huang + Benioff vs. market fears), which has direct implications for SaaS pricing, seat counts, and ISV strategy.
  • It legitimizes the open-model + proprietary-data path for enterprise AI, a direct competitive positioning against frontier-lab API dependence.
Evidence

CONFIRMED

10 sources · 39 min read
Story identity
  • Story ID: S38
  • Title (discovery): "Jensen Huang's Dreamforce remarks: agents will 'program themselves' and rewrite enterprise software"
  • Organization: NVIDIA (with Salesforce as co-announcer)
  • Category: enterprise
  • Event date: 2026-09-15 — Dreamforce 2026 Day 1 keynote, Moscone Center, San Francisco (Jensen Huang on stage with Marc Benioff)
  • Announcement date: 2026-09-15 (Koa announcement, same keynote)
  • Article dates: NVIDIA blog 2026-09-15; Salesforce press release 2026-09-15; CNBC/TechCrunch/TechRadar 2026-09-15; QZ 2026-09-16; FT 2026-09-16; TechChautari 2026-09-16

Evidence status

  • Overall story: CONFIRMED. The keynote, the Huang–Benioff joint appearance, the Koa announcement, and the safety divergence between Huang and Anthropic CEO Dario Amodei are corroborated by multiple independent outlets (CNBC, TechCrunch, TechRadar, QZ) plus the two primary company sources.
  • Title phrasing caveat (EARLY RESEARCH): the discovery title says agents will "program themselves" and "rewrite enterprise software." No direct quote matching that phrasing was located in any accessed source during this research. The verified statements support an agentic layer on top of software thesis ("This is going to be a layer on top of software, this new layer is going to be agentic, and it's going to make the software so much better" — Huang via TechRadar). The "program themselves" claim is treated as INTERPRETATION of Huang's remarks, not a traced quote, and is flagged as such in this report.
  • Discovery evidence_status: CONFIRMED, confidence High. Impact scores (discovery): importance 6, technical 7, enterprise 8, developer 7, strategic 7; content/training/consulting potential 8.
✓

What happened?

🎓 For Explorer
  • FACT: On 2026-09-15, NVIDIA CEO Jensen Huang joined Salesforce CEO Marc Benioff on the Dreamforce keynote stage at Moscone Center, San Francisco. CNBC reported roughly 12,000 people attended the event; OpenAI CEO Sam Altman also appeared at Dreamforce, and Anthropic CEO Dario Amodei shared the stage around the same event, where he and Huang publicly diverged on AI safety (CNBC).
  • FACT: Salesforce and NVIDIA announced Koa, described by Salesforce as its first CRM reasoning model for Agentforce, built on NVIDIA Nemotron (Salesforce press release; corroborated by TechCrunch).
  • COMPANY CLAIM (attributed): Huang argued AI has become "an infrastructure layer across the planet," that "every company, every enterprise, every country would become an AI company," and "It's going to be an agentic enterprise." He closed with "Engage AI. Don't get left behind." (NVIDIA blog, 2026-09-15).
  • COMPANY CLAIM (attributed): Huang rejected the "SaaSpocalypse" thesis: "Don't give up on the technology... I was the first CEO to come out not in the software industry to say the end of software is nonsense." He and Benioff agreed AI "is going to be a layer on top of software" that is "agentic" and "going to make the software so much better" (TechRadar, 2026-09-15).
  • COMPANY CLAIM (attributed): On safety and regulation Huang said "Safety is an engineering problem, not a legal one," and "The market forces are already there. We don't need any new laws. We don't need new regulations." (TechCrunch, 2026-09-15). The NVIDIA blog similarly recorded "Safety is paramount in a lot of ways. It's job one... However, safety is an engineering problem..."
  • INDEPENDENT EVIDENCE: QZ (2026-09-16) characterizes Huang as rejecting AI-slowdown calls at Dreamforce. FT headline (2026-09-16, paywalled): "Nvidia and Meta bosses reject efforts to co-ordinate AI slowdown."
Δ

What changed?

  • INTERPRETATION (primary): The week's defining enterprise claim — per discovery why_it_matters — is that autonomous agents are positioned as "the new category of software." Huang and Benioff reframed the SaaSpocalypse narrative: AI does not end software; it adds an agentic layer on top of it.
  • FACT (framing change): Salesforce's flagship AI asset moved from a general assistant to a domain reasoning model — first CRM reasoning model for Agentforce — anchoring agent intelligence in proprietary enterprise data rather than a frontier lab's API.
  • COMPANY CLAIM (structural change implied): Enterprise software investment shifts from "tools that humans operate" toward "agents that act," with NVIDIA providing the open foundation model (Nemotron) and Salesforce providing the enterprise reasoning layer.
  • The discovery's what_changed ("software engineering will shift from humans writing code to agents self-modifying systems") is INTERPRETATION/paraphrase; no accessed source contains a verified Huang quote about agents self-modifying code or rewriting enterprise software.
↔

Before → Change → After

🎓 For Explorer
  • Before: Enterprise AI was largely assistants/co-pilots that recommend; discourse was split between "SaaSpocalypse" doom (AI coding agents make software cheap to build) and a coordinated-slowdown debate; CRM automation was rules/workflow-driven within Salesforce and rival platforms.
  • Change: At Dreamforce 2026, NVIDIA's CEO and Salesforce's CEO jointly declared AI an infrastructure layer that makes software better, not obsolete; Koa brings reasoning (multi-step planning, tool use, action-taking) inside the CRM runtime; Huang publicly split from Amodei on whether safety requires coordination/regulation, and (per FT headline) joined Meta's Zuckerberg in rejecting coordinated slowdowns.
  • After: Enterprises are pushed toward "agentic enterprise" planning — agents as workers inside platforms like Agentforce; expect a wave of domain reasoning models built on open models (Nemotron) post-trained on proprietary enterprise data, competing with general frontier APIs; expect continued public tension between industry leaders on safety vs. speed.
⚙

How it works

  • COMPANY CLAIM (Salesforce press release, 2026-09-15): Koa was built by post-training NVIDIA Nemotron 3 Super — an open NVIDIA model — using a proprietary synthetic dataset modeled on enterprise knowledge drawn from nearly three decades of Salesforce CRM deployments.
  • COMPANY CLAIM: Koa is designed to help agents reason through complex, multi-step workflows and use the right tools at the right time, i.e. plan, select tools, and take action rather than merely generate text.
  • COMPANY CLAIM: Models run within Salesforce infrastructure (data-governance messaging: enterprise knowledge stays in the CRM platform), distinct from calling frontier labs' hosted APIs.
  • INDEPENDENT EVIDENCE (TechCrunch analysis, 2026-09-15): The strategic point is that Salesforce pairs "27 years of CRM intelligence" with NVIDIA's open models to produce a reasoning model AI labs should fear — a proprietary-data play, not a compute-play alone.
  • Note: the mechanism by which agents would "program themselves" or "rewrite enterprise software" is not specified in any accessed source; current verified mechanism is agentic tool use/reasoning over enterprise data.
!

Why it matters

🎓 For Explorer
  • Enterprise impact is the highest dimension (discovery: enterprise_impact 8): CRM is the front office of most companies; a reasoning model that can act (update opportunities, route cases, schedule follow-ups) inside Agentforce changes what "CRM software" does.
  • It reframes the SaaSpocalypse debate with CEO-level authority on both sides (Huang + Benioff vs. market fears), which has direct implications for SaaS pricing, seat counts, and ISV strategy.
  • It legitimizes the open-model + proprietary-data path for enterprise AI, a direct competitive positioning against frontier-lab API dependence.
✦

What became possible?

🎓 For Explorer
  • COMPANY CLAIM / INDEPENDENT EVIDENCE (mixed): CRM agents that reason through multi-step workflows (planning + tool use) rather than single-shot Q&A.
  • COMPANY CLAIM: Enterprises can benefit from reasoning models trained on "27 years of CRM intelligence" without sending their CRM data to a general-purpose frontier API (governance benefit).
  • INTERPRETATION: Vendors began publicly claiming that the next software category is agents — a claim now available to enterprises as a procurement/strategy framework ("agentic enterprise," "know everything, do anything").
◎

Implications

Technical

  • COMPANY CLAIM: post-training an open foundation model (Nemotron 3 Super) on proprietary synthetic enterprise data is presented as a viable route to domain-grade reasoning — a counter-model to frontier-API-only strategies.
  • COMPANY CLAIM: Koa is positioned against Salesforce's CRM benchmark (real-world tasks such as updating an opportunity, routing a case, scheduling a follow-up) — COMPANY CLAIM, pending independent benchmark verification.
  • INTERPRETATION: If this succeeds, expect enterprise reasoning benchmarks to fragment by domain (CRM, ERP, HR) rather than a single general leaderboard.
  • FACT (general context): Nvidia simultaneously extends "Missionforce" positioning — bringing NVIDIA open models and accelerated computing into Salesforce's platform (press release).

Developer

  • INDEPENDENT EVIDENCE (TechRadar/TechCrunch reporting): Developers are told to build on/with agents rather than only writing application logic; the "end of software" fear is reframed as an added layer on top of software.
  • INTERPRETATION: Skills shift toward agent orchestration, tool/API design, evaluation and guardrails for agent actions; headless/API-first integration becomes more central (e.g., the Anthropic-via-Claude/Slack pattern noted in The Register's earlier "headless" coverage is context, not this story's evidence).
  • EARLY RESEARCH: No developer-facing spec, SDK, or self-modification capability was published with this announcement; the "agents rewriting software" framing has no shipped mechanism yet.

Enterprise

  • INDEPENDENT EVIDENCE (TechCrunch): The most enterprise-relevant threat/opportunity is that proprietary CRM data becomes the moat — enterprises should evaluate who controls the reasoning layer over their data.
  • COMPANY CLAIM: Koa runs inside Salesforce infrastructure, directly addressing enterprise governance concerns about enterprise data flowing to external model providers.
  • INTERPRETATION: Pricing/packaging shifts from seats toward agent actions and metered usage (Flex Credits context at Salesforce) — procurement should model agent-workload costs, not just license costs.
  • INTERPRETATION: "Engage AI. Don't get left behind" translates into an urgency signal for enterprise AI roadmaps.

Strategic

  • INDEPENDENT EVIDENCE: The Huang–Amodei safety divergence (CNBC) and the Huang/Zuckerberg rejection of coordinated slowdowns (FT headline) mark a visible strategic split among AI leaders — speed-and-market vs. coordination-and-safety camps.
  • COMPANY CLAIM/INTERPRETATION: NVIDIA positions itself beyond silicon — open models (Nemotron) as an enterprise-software platform layer in alliance with Salesforce; a direct flank against OpenAI/Anthropic enterprise ambitions.
  • INTERPRETATION: The "agentic layer" framing is strategically convenient for both NVDA (sells more accelerated compute) and CRM (sells more agentic seats/credits) — acknowledge this incentive when reporting.
⚠

Risks & limitations

Risks
  • COMPANY CLAIM (unverified substance): Koa benchmark/quality claims are vendor-published; no independent benchmark verification exists yet — risk of overstating capability.
  • INTERPRETATION: Agent action errors in CRM (wrong tool call, incorrect record update, cascading automation) are a real operational risk; no independent safety evaluation of Koa was published.
  • INDEPENDENT EVIDENCE (context): Huang's "no new laws/no new regulations" stance runs contrary to calls from Amodei and others for safety coordination — reputational/regulatory risk if incidents occur.
  • FACT: Salesforce sustained a global outage on 2026-09-16, as thousands attended Dreamforce (The Register, 2026-09-16) — reliability of the hosting platform is a live concern for agent-heavy workloads (context; not part of this story's core claims).
Limitations
  • No direct quote located for the headline phrase "program themselves" / "rewrite enterprise software" — that element of the story remains INTERPRETATION/EARLY RESEARCH and is not independently verified.
  • Full keynote transcript/recording was not accessed; quotes are as reported by NVIDIA blog and independent press (CNBC, TechCrunch, TechRadar), which may compress wording.
  • FT article is paywalled — only the headline was confirmed; the Amodei counter-argument is only partially covered by CNBC/TechRadar.
  • The Register was listed in discovery's independent sources but no Dreamforce keynote article was locatable at theregister.com during research; the The Register pieces referenced are context only.
  • Koa pricing, GA timing, and API availability were not confirmed in accessed sources.
?

Open questions

  • When does Koa ship generally, at what price, and under what metering?
  • Will independent benchmarks (e.g., Salesforce's CRM benchmark) confirm the vendor's claims?
  • What exactly did Huang say, per the official transcript, about software evolution — did he use any "programs itself" style formulation?
  • How will OpenAI/Anthropic respond on enterprise reasoning, and does the Huang–Amodei safety split harden into policy positions?
  • Do enterprise customers actually move from seat-based SaaS to agent-based consumption, or is that messaging?
↗

What happens next?

🎓 For Explorer
  • PREDICTION: Salesforce ships Koa generally within the next 1–2 quarters, followed by independent benchmark evaluations that will moderate or confirm vendor claims.
  • PREDICTION: Competitors (Microsoft, OpenAI, Anthropic, ServiceNow) respond with their own domain-reasoning or enterprise-agent plays; the "CRM reasoning" space becomes contested within 2026.
  • PREDICTION: The Huang–Amodei safety divergence becomes a recurring news theme; expect policy bodies (EU AI Office, UK, US Congress) to cite the split in hearings.
  • WATCH: Koa pricing/metering, Salesforce Flex Credits economics, and whether "agents as software" changes seat-based SaaS revenue models.
★

Editorial takeaway

🎓 For Explorer
  • The verified story: at Dreamforce 2026, NVIDIA's CEO and Salesforce's CEO jointly declared AI an agentic layer on top of enterprise software, unveiled a CRM reasoning model (Koa) built on NVIDIA's open Nemotron 3 Super and decades of Salesforce data, and Huang publicly rejected regulation and slowdown coordination, splitting from Amodei on safety.
  • The discovery headline's "program themselves" phrasing must not be presented as a quote — the verified, attributable formulations are "agentic layer," "agentic enterprise," "know everything and do anything," and "make the software so much better." Report those; flag the stronger version as interpretation.
  • Balance: NVIDIA/Salesforce claims are company claims with strategic incentives (compute sales; agentic seats); independent confirmation covers the event, the product's existence, and the safety divergence, not yet the product's quality.
A used engineering workbench with tools and a checklist sits beside a substantial law library of bound volumes with a closed case put away on a high shelf.
⌘

Lab: NO-LAB

≡

Research sources

Primary Sources (2)
Primary
Salesforce press release — "Announcing Koa: Salesforce's First CRM Reasoning Model, Built on NVIDIA Nemotron"Koa product details — post-trained from NVIDIA Nemotron 3 Super with a proprietary synthetic dataset modeled on ~27 years (nearly three decades) of CRM deployments; reasoning over multi-step workflows with tool use; runs in Salesforce infrastructure; Salesforce CRM benchmark reference; Missionforce; direct Jensen Huang quote. — Primary company source (COMPANY CLAIM for capabilities/benchmarks).Date: 2026-09-15
Visit source ↗
Primary
NVIDIA blog — "'Now We Can Know Everything and Do Anything,' Jensen Huang Says at Dreamforce" (author: Brian Caulfield)Huang's keynote remarks — "infrastructure layer across the planet," "every company... would become an AI company," "agentic enterprise," "Engage AI. Don't get left behind," safety-as-engineering-position, and the Koa/Nemotron announcement framing. — Primary company source (COMPANY CLAIM for quotes/claims; FACT for event occurrence).Date: 2026-09-15
Visit source ↗
Independent Sources (7)
Independent
Financial Times — "Nvidia and Meta bosses reject efforts to co-ordinate AI slowdown"Headline only (paywalled): Huang alongside Meta's Zuckerberg rejecting coordinated AI-slowdown efforts. — Independent source, headline-level confirmation only; full text not accessible.Date: 2026-09-16
Visit source ↗
Independent
QZ — "Jensen Huang rejects AI slowdown calls at Dreamforce 2026"Independent characterization of Huang rejecting AI-slowdown calls; links safety debate to Anthropic/OpenAI context. — Independent source.Date: 2026-09-16
Visit source ↗
Independent
TechRadar — "'We're going through a new industrial revolution': Nvidia CEO Jensen Huang and Anthropic CEO Dario Amodei share differing views on AI at Dreamforce 2026""New industrial revolution" framing; Huang vs. Amodei differing views on AI at the same event. — Independent source for the industrial-revolution theme and CEO divergence.Date: 2026-09-15
Visit source ↗
Independent
TechRadar — "Salesforce CEO Marc Benioff says the SaaSpocalypse is 'crazy nonsense' - and Jensen Huang agrees with him"Direct quotes — "Don't give up on the technology... I was the first CEO to come out not in the software industry to say the end of software is nonsense." and "This is going to be a layer on top of software, this new layer is going to be agentic, and it's going to make the software so much better." — Independent source for attributed quotes on the agentic-layer framing; basis for the "program themselves" interpretation caveat.Date: 2026-09-15
Visit source ↗
Independent
TechCrunch — "Salesforce and Nvidia's new reasoning model is everything the AI labs should fear"Independent analysis of Koa's strategic meaning — proprietary CRM data moat ("27 years of CRM intelligence") combined with NVIDIA open models; competitive position vs. AI labs. — Independent analytical source.Date: 2026-09-15
Visit source ↗
Independent
TechCrunch — "We don't need AI regulation — leave safety to us, Nvidia's Jensen Huang says"Huang's direct quotes on regulation/safety: "Safety is an engineering problem, not a legal one." and "The market forces are already there. We don't need any new laws. We don't need new regulations." — Independent source for attributed quotes and the no-regulation stance.Date: 2026-09-15
Visit source ↗
Independent
CNBC — "Nvidia and Anthropic CEOs diverge on AI safety at Dreamforce"Huang–Amodei safety divergence on the same stage; OpenAI's Sam Altman also appearing; ~12,000 people at Moscone Center; independent confirmation of the event. Independently verifies the event itself and the safety split. — Independent source (INDEPENDENT EVIDENCE).Date: 2026-09-15
Visit source ↗
Secondary Sources (1)
Secondary
TechChautari — "Benioff and Huang dismiss 'SaaSpocalypse' fears at Dreamforce 2026"Secondary corroboration of the Benioff–Huang SaaSpocalypse rebuttal and the joint stage appearance. — Secondary corroboration; not used for primary quotes.Date: 2026-09-16
Visit source ↗