ClarityBeta

Product & Strategy

The Product Desk

9 min read · 1,838 words · 3 moves to watch
The Signal

OpenAI's free first Dot reaches 4,000+ apps before your AI-teammate sales call.

Each agent gets one cloud computer of its own and works from three surfaces buyers already use: ChatGPT, Slack and Teams. That means the generic delegation pitch on your sales call is answered before you dial. The open ground is narrower: enterprise accounts, where an admin has to switch the beta on, and any job a generalist can't reach or safely reverse.

In Play

  1. OpenAI Bundles an Always-On Agent

    At DevDay, OpenAI launched Dots, per TheSequence. Dots are always-on agents that run on their own cloud computer, connect to 4,000+ apps and work across ChatGPT, Slack and Teams. The first Dot costs nothing extra on Pro and Business Premium plans. GPT-6.1 Sol launched at $2 input and $10 output per million tokens. Any horizontal 'AI teammate' feature you sell now competes with an agent your buyer already pays for, inside their chat tools.

    Ask Clarity
    Sign in to ask more about this

  2. Outside Checkers Beat Bigger Models

    In an NVIDIA–KAIST study reported by TheSequence, a small agent's first-try success rose from 50.00% to 68.03% when a stronger model checked each action before it ran. When the agent checked its own work, it reached only 54.76%. Separately, Google-led RRSI research found that harness tweaks (edits to the prompts and tools around a model) tuned on one eval set fade on new tasks. On your agent roadmap, an outside verifier and a held-out test set now do more for quality than swapping models.

    Ask Clarity
    Sign in to ask more about this

  3. Uber Ships Agent Tools Switched Off

    Per The ML Engineer, Uber routes every AI agent through one gateway that covers 800+ MCP servers and 5,000 tools. MCP is the protocol agents use to call software tools. Every tool starts disabled until someone enables it. Uber's scaling problems were permissions, context bloat and cost, not model quality, so your tool catalogue and delegation rules are now product spec. Separately, Anthropic's Alex Palcuie rates Claude for incident response as only a 'maybe.'

    Ask Clarity
    Sign in to ask more about this

  4. Tomorrow's Cheap Compute Is Still Unbuilt

    Per The Information, Nscale has filed to go public at a reported $35 billion target valuation. Nscale is a neocloud, a company that rents out AI chips. Nearly all of its $103 billion in contract commitments depends on 12 data centers that are not yet built or fully financed. Its Anthropic, Microsoft and Figure AI deals ramp in 2027–2028, the window your multi-year AI cost assumptions cover. The Bear Cave adds that Nvidia is in early talks to insure lenders against smaller GPU clouds defaulting.

    Ask Clarity
    Sign in to ask more about this

Deep Dives

  1. Your Top User Jobs Now Have a Bundled Competitor in Slack

    When the agent comes inside a plan your buyer already pays for, defensibility depends on what a generalist can't reach and on what your own features cost to keep running.

    An ops lead on Business Premium who wants a vendor sheet watched does not shop for an "AI teammate." She types @ChatGPT in Slack. Per TheSequence's DevDay recap, Dots are rolling out now to Pro and Business Premium users in…

    2 action items

    ●

  2. Agent Gains Came From the Checker, Not the Model

    Three results move the verification step and the held-out test onto the product roadmap. Most teams still file both under QA overhead.

    When an agent misfires, the cheapest fix teams reach for is a second prompt that asks it to check its own work. NVIDIA and KAIST's Mid-Harness study tested that fix directly, and it mostly failed. The researchers also distilled the…

    3 action items

    ●

  3. Uber Ships 5,000 Agent Tools Switched Off, and That's the Spec to Copy

    When tools are generated faster than anyone can review them, the default setting and the on-whose-behalf rule decide what your agent is able to break.

    The most telling detail in Uber's write-up is how the tools get made. A workflow called AutoCrawler continuously scans Uber's API definitions and generates MCP servers from them, and an LLM writes the tool descriptions. That is how a company…

    3 action items

    ●

The edition continues

Take the signal into the room.

Sign up or log in to read all 3 deep dives in full, plus the final take.

Read the full edition

Continue with LinkedIn