010 AI For Operators

Ana Silva, Chief of Staff at Section, shares how operators can drive deeper, faster AI adoption across their organizations.

• Partnership-driven upskilling: Section moves teams from AI-curious to AI-proficient with an AI coach and tailored change-management playbooks.

• ROI gap: Firms that only buy LLM seats see weak returns; Section insists on an AI manifesto, usage policy, and workflow redesign before licenses.

• New-tech rollout ≠ SaaS launch: Probabilistic outputs and job-security fears demand dedicated champions, iterative training, and psychological safety.

• LLMs as board-meeting sparring partners: Ana role-plays aggressive and risk-averse directors to surface tough questions before leadership reviews.

• Section Expert knowledge base: Claude drafts project briefs, partner proposals, and content in seconds using company OKRs, templates, and exec user guides.

• Experimentation culture: Weekly AI lunch-and-learns plus ‘shout-outs’ and ‘anti-shout-outs’ publicly reward explorations and own misfires.

• Manager pull effect: Direct leaders who model AI usage correlate with higher team adoption across Section’s client benchmark data.

What It Does

Dust combines cross-platform knowledge search with easy AI Agent creation.

Why Ops Leaders Should Care

AI Agents are exciting, but the real key to extracting value from AI in the enterprise is maximizing adoption. Dust customers say this is where the platform excels - with an intuitive interface, powerful integrations, and ability to collaborate across your full team, Dust should be less intimidating than other tools.

Key Features (Pros & Cons)

Pros

• Proprietary “smart search” is the best on the market, preserving document structure and making retrieving the info you need very easy across lots of sources - users report that Dust’s RAG is “far more effective than Microsoft Copilot’s or ChatGPT’s”

• Easy for non-technical users to get started: a lot of the technical jargon and tasks are abstracted into a UI that pretty much anyone can get the hang of

• Simple to customize: create multiple, single-function agents and choose which model you’re using on a tool-by-tool or query-by-query basis

• Multi-function capabilities: string together multiple tasks with a single agents

• Interact with Dust agents where you work: chat with them on Slack or use their Chrome Extension

• Strong customer service: the helpful community Slack and responsive customer service are consistent themes across reviews

Cons

• Simple UI leans more ‘prosumer’ than sophisticated enterprise user

• Isn’t powerful ‘out of the box’: like most systems of this kind, it’ll require a fair amount of effort to connect all your data sources, customize your agents, and make sure permissions are set up correctly

• Data sync is sometimes slow or incomplete: some customers mention encountering issues with access to certain Notion databases, making their platform’s output less useful

• Cost: while a 14-day free trial is available, the Pro plan is $29/user/month. While this isn’t unreasonable (and includes usage of the underlying models, which can add up), for teams at small companies, this could be too much to stomach.

An Operator’s Perspective

Agents have been the hot topic of AI for a while now, so I figured it wouldn’t hurt to check out the Sequoia-backed, Paris-based agent platform. The platform was easy to onboard to, though I didn’t go to the extent of connecting all of my company’s tools to it.

The agent creation flow will be intuitive to anyone who has created a custom GPT on ChatGPT.

Which leads me to my real take here: it’s difficult to see how this is substantially better than ChatGPT (though user reviews claim that the search/access to tools is much better). While there are important reasons to buy an enterprise skin on a multi-model infrastructure base (collaboration, security, etc.), calling this an “agent” platform seems like a bit of a stretch.

Also noteworthy, while Sequoia led Dust’s Seed and Series A, they haven’t announced a new round since mid-2024. In the hottest market ever for AI startups, that’s interesting…

Other Options

Bottom Line

If you’re looking for a simple but powerful tool to spin up agents and collaborate across your team, and you’re not scared of the price point, Dust could be worth a try.

Practical

• What are sub-agents? The gang at Every discusses one of Anthropic’s most recent releases - sub-agents - what they’re good at, and where they fall short today.

• Replacing software with custom AI tools: 12 examples of people who have rolled their own tools, launched quickly, and saved lots of money in the process.

• The evolution of AI Agents: The CEO of Dust and CEO of LangChain discuss the capabilities of agents today and where they’re headed.

• Launch a brick-and-mortar business with AI (in your spare time): The founder of Groupon and a GP at Spark Capital launched a board game social club in Berkeley, leaning heavily on AI through the whole process.

Perspectives

• Yes, AI models have distinct personalities: A new paper from Anthropic identifies ‘persona vectors’ - the patterns of activity that determine whether a model is curious, sycophantic, direct, or other personality traits. Companies (and consumers and governments) are likely to demand more transparency from big AI companies on how they’re shaping the personalities of their latest releases.

• Why I believe in AGI (again): Alexey Guzey, an OpenAI employee, discusses how he went from doubting to believing.

• AI is polytheistic, not monotheistic: The always-interesting Balaji gives us ten thoughts on AI that he finds personally economically useful.

• AGI and the economy: Dwarkesh, Noah Smith, and Erik Torenberg debate how impactful AI is on the economy today, whether AGI will bring massive inequality, and much more

News

• The Swedish Prime Minister gets advice from AI: He “often” uses it for a second opinion and compared his usage to that of doctors who also use it to obtain a different perspective. It’ll be fascinating to see which world leaders use AI to signal how forward-thinking they are vs. the ones that double down on anti-AI rhetoric to mollify the anti-tech crowd and those who fear job displacement.

• OpenAI released their open source model: long expected, but this has important tech and economic implications (cheaper and can be run locally on your laptop or phone, depending on the model, according to Altman) as well as geopolitical implications (Chinese companies have been competing hard with each other to release the best open source models - and succeeding)

• Anthropic released Opus 4.1: an upgrade to Opus 4.0, most relevant to those who use Claude Code.

Thanks for reading,

Tom Guthrie

If you liked this edition of AI for Operators, share it with another operator you know!