Koko Daily · Wednesday, September 30, 2026

Koko: It's Wednesday, September thirtieth, I'm Koko, with Max and Sam — let's blitz it.

Max: OpenAI's DevDay dropped over twenty products in one day, headlined by always-on agents called Dots.

Sam: And a cheaper GPT six point one Sol model, priced at a fifth of the old flagship.

Koko: But Sam Altman also said OpenAI won't go public until it can make, quote, confident safety claims.

Max: Meanwhile Anthropic's IPO prospectus confirms the number — two trillion dollar valuation target, against a forty-two billion dollar net loss.

Sam: Meta launched an Enterprise Platform, and days later its Muse agent leaked a user's home address to a stranger.

Koko: IBM Consulting says AI agents are shrinking project teams and raising the skill bar for who's left.

Max: BCG says AI and tech services now top forty percent of its revenue.

Sam: And a GFT survey finds over eighty percent of CIOs have canceled an AI pilot because of legacy systems. Big one, let's dig in.

Koko: Okay, this matters right now because it's OpenAI basically saying two contradictory things on the same day.

Max: Right — twenty-plus products, an app store inside ChatGPT, always-on agents called Dots that just... keep running. That's a company sprinting.

Sam: And then Altman turns around and says, actually, hold on the IPO, we're not ready to vouch for safety yet.

Koko: Which is a weird flex, honestly. Ship faster than anyone, but tell Wall Street to wait.

Max: It's not that weird if you think of it as risk transfer. Public shareholders get a safety promise. Enterprise customers just get Dots on a Tuesday.

Sam: [chuckles] So the free tier is the beta test and the IPO prospectus is the marketing copy.

Koko: Harsh, Sam, but go on.

Sam: I mean, always-on agents are exactly the category that's been breaching Medicare portals and UN sites lately. Dots isn't emerging in a vacuum.

Max: And the Sol model — one-fifth the cost of the old flagship — that's not a safety story, that's OpenAI trying to out-cheap Anthropic before the IPO clock runs out.

Koko: So which is it — is delaying the IPO genuine caution, or a company that just doesn't want a lawsuit filed the same week as its prospectus?

Max: Probably both. Genuine caution is also just good legal hygiene when your prospectus disclosures could become Exhibit A.

Sam: The tell for me is the Agents API and Spaces launching anyway. If you're truly worried about safety claims, you don't also open an app marketplace the same week.

Koko: Right, the marketplace ambition is the part nobody's talking about enough — that's OpenAI trying to become the platform, not just the model.

Max: Which puts it on a collision course with Meta's enterprise push and Microsoft's Copilot super app. Everyone wants to own the workspace, not just the API call.

Sam: And whoever wins that owns the distribution. The safety delay is a footnote next to that fight.

Koko: A footnote that could still blow up the IPO timeline. We'll see who blinks first.

Koko: This one matters because Meta picked the worst possible week to ask businesses to trust it with their data.

Sam: Launch the Enterprise Platform, hire the MongoDB CEO to run it, big pitch — and days later Muse hands a stranger someone's home address.

Max: That's not a rounding-error bug, that's the exact fear every enterprise buyer has about agentic AI — it leaks what it shouldn't, to whoever's asking.

Koko: And Meta doesn't exactly walk in with a trust surplus on privacy to begin with.

Sam: [laughs] No, this is maybe the one company where that headline just... writes itself.

Max: Here's the thing though — technically it's fixable. Patch the leak, add guardrails, move on. The real damage is reputational timing.

Koko: Because now every CIO evaluating Meta's platform has a built-in objection in their back pocket.

Sam: Right, and that's happening in the same week OpenAI's agents are getting flagged for unauthorized site hits and Meta itself patched a separate Muse zero-day for hijacking.

Max: So it's not just Meta's problem, it's an industry pattern — ship the agent, find the hole in production, patch under pressure.

Koko: Which raises the actual question: is anyone's enterprise agent platform ready, or are we all just watching who gets caught first?

Sam: Meta just volunteered to go first this week. That's rough branding for a launch.

Max: The MongoDB CEO hire is interesting context, though — that's an infrastructure and data-governance mind, not just a product person. Maybe that's the tell they know the trust gap is the real product.

Koko: Building the plane while flying it and also live-streaming the emergency landing.

Sam: [chuckles] I'd watch whether any of Meta's early enterprise customers pause deployment pending an independent security review. That's the real signal, not the apology blog post.

Koko: This matters right now because it's the clearest data point yet that AI isn't just changing consulting delivery, it's changing consulting economics.

Max: IBM Consulting is saying it out loud — agents absorb the routine work, teams get smaller, and the bar for who's on them goes way up.

Sam: That's the traditional pyramid model in trouble. Junior analysts doing grunt work to learn the business — that's the layer getting automated first.

Koko: And it's not theoretical anymore — BCG says AI and tech services are over forty percent of its revenue now, with AI-specific work alone near a quarter.

Max: That's not a pilot number, that's a structural shift in the P and L. When a firm that size reweights revenue like that, everyone downstream has to react.

Sam: Accenture's pushing the same story from a different angle — one point four times the margin, two point two times the shareholder returns for managed-services adopters over staff-aug billing.

Koko: So the pitch to clients is shifting from, we'll staff your project, to, we'll deliver your outcome.

Max: Which sounds great until you realize outcome-based pricing means the firm keeps more of the value AI creates instead of passing savings to the client.

Sam: [laughs] Right, nobody's cutting invoices in half out of generosity.

Koko: But here's the tension — Sloan Review's also reporting in-house teams are pulling billable work back from outside providers because they've got their own generative AI now.

Max: So the squeeze comes from both directions — clients need fewer hours from you, and the hours you do sell have to come from senior, specialized people who cost more.

Sam: Which is exactly why KPMG's throwing money at the accounting talent pipeline. That's a hedge against exactly this compression hitting entry-level roles.

Koko: So the firms betting right now are the ones restructuring staffing ratios before the market forces it on them.

Max: Correct. The ones still selling by the hour in twenty twenty-seven are going to have a very uncomfortable renewal conversation.

Sam: Cognizant deployed agentic claims processing for health plans — routine work automated, staff freed for exceptions, a real production example, not a pilot.

Koko: Alibaba's going full agent-native at Apsara, spanning chips to infrastructure to global operations — a real full-stack bet, not just cloud services.

Max: Forrester's flagging a minimum viable data framework for agentic supply chains — basically, don't let your agents run before your data's harmonized.

Sam: And Telefonica's weighing a sale of its Venezuelan unit, a reminder that plain old portfolio pruning is still happening alongside all this AI noise.

Koko: Okay, let's recap. OpenAI shipped over twenty products at DevDay but is delaying its own IPO until it can vouch for model safety.

Max: Anthropic's prospectus confirms a two trillion dollar valuation target against a forty-two billion dollar net loss, still barreling toward going public.

Sam: Meta launched its Enterprise Platform right as its Muse agent leaked a user's home address — trust gap, meet timing problem.

Koko: And consulting's leverage model is visibly inverting — IBM, BCG, and Accenture all pointing at smaller teams, higher skill bars, and outcome-based pricing.

Koko: Now, if you're sitting in professional services or private equity, here's the lens for today.

Max: Think about this: every one of these labs is racing to own the agent workspace while quietly admitting they can't fully control what the agents do. That's the exact risk profile PE diligence teams need to be pricing into portfolio companies buying agentic tools right now — not next quarter.

Sam: And for the firms themselves — if BCG's at forty percent AI and tech revenue and IBM's shrinking teams, ask yourselves plainly: is your staffing pyramid built for twenty twenty-three, or for what's actually happening in twenty twenty-six?

Koko: Do this next — audit your billable-hour engagements this month for exposure to in-house AI substitution, the way Sloan Review flagged. If clients can do it themselves cheaper, you need a new pitch before they figure that out.

Max: And for PE specifically — when you're marking portfolio company value, ask whether that valuation already assumes AI-driven margin expansion or compression, because pretending it's neutral is the mistake right now.

Sam: Open question worth sitting with: when a services firm's value shifts from headcount to agent-governed outcomes, who actually owns the IP created in that engagement — the firm, the client, or the model vendor?

Koko: Watch for at least one major consulting firm to publicly cut junior analyst hiring targets for next year's class within the next month, directly citing agent leverage.

Max: And watch for OpenAI's IPO timeline to slip into next year — Altman's safety language today reads like a company buying itself at least two more quarters.

Sam: Longer view — within the year, expect at least one major PE firm to formally adjust portfolio valuation methodology to explicitly price in AI-driven repricing, the way PwC's already flagging.

Koko: So much to chew on. If you want the full brief, the sourcing, and the lens built for your specific role, go check us out.

Max: Seriously, bookmark it — it updates daily and it's sharper than anything we just crammed into fifteen minutes.

Koko: Head to koko knows dot A I. We'll see you tomorrow.