Last week the founder of Sentry went viral on X with his insight on specialized vs universal agents for work:
“Vendor-specific chatbots are broken by design. The Sentry agent, the Linear agent, and any others you might have in Slack are fine for some point situations, but agents with generalized access outperform them in every single scenario.” — David Cramer, founder of Sentry, on X
The data supports Cramer’s experience. Accountants using vendor-specific AI are 21% more productive. Similarly, salespeople win 9.4% higher revenue deals and lawyers save 10 hours per week. The largest impact is on developers. 88% say a coding agent helps them ship faster and feel more fulfilled with their work.
So if every individual contributor is winning with AI, why are companies failing with it? According to consulting firm, PWC:
"Despite widespread experimentation, only one-in-eight (12%) CEOs say AI has delivered both cost and revenue benefits. Overall, 33% report gains in either cost or revenue, while 56% say they have seen no significant financial benefit to date."
It's a productivity paradox. AI agents are helping individual contributors produce results faster, but companies are not reporting an increase in revenue.
Why specialized AI agents failed
The first AI agent wave was defined by legacy SaaS apps trying to retrofit intelligence into the products they already sold. GitHub launched Copilot, and developers got a coding agent. Then Salesforce rolled out Agentforce, and sales and support got customer agents. Each tool made a specific function faster, but none of them were designed for how teams work together.
As Cramer explained in his tweet, specialized agents create local wins without changing company throughput. This is because vendor-specific agents exacerbate the classic local maxima problem. Agents helped each team climb its own hill faster, but it did not help company leaders find the highest-value mountain to collectively scale.
Cramer's company discovered the solution to this problem by accident. Sentry built an internal Slackbot, connected it to systems like GitHub, Linear, Notion, and even Sentry itself. By the time Linear added code search capabilities to their agent, Cramer’s team already had a multi-purpose agent that could search code and push the information wherever it needed to go. Sentry tested bot solutions and found that what they built could outperform the vendor-specific agents.
That scenario describes why we've been building a universal agent that every team can use.
The universal AI agent for work
Most company work is cross-functional, so a company will only reach its full potential when data, decisions, and execution flow between teams without friction.
That requires more than an AI agent inside each team’s favorite app. You need a single agent that can connect to company data and systems and can access shared knowledge and workflows across all teams - a shared company brain.
That is what makes a universal agent powerful. While a specialized agent is handcuffed to its parent product and associated use cases, a universal agent connects to any system and can work across surfaces
A universal agent starts with the outcome a team wants to achieve and can creatively work through problems. The agent can follow a question from Slack to HubSpot, discover an anomaly in product telemetry, track a root cause to a Linear ticket, and design a report for the teams making the decision.
Adapt is a universal AI agent that gives teams one shared intelligence layer for company data, multi-app actions, and reusable work. Teams can draw on the same company knowledge instead of working from scattered docs and data. The agent can take action across systems instead of getting trapped inside one SaaS tool. And when someone creates a workflow, app, dashboard, report, or skill, the rest of the company can build on it instead of starting from scratch.
Why universal agents are now possible
In the past, specialized agents worked well because their harnesses planned around a limited set of known use cases.
But that was a workaround for the model limits at that time. In 2025, AI capability dramatically increased agency, reasoning, and context handling. Or said plainly, models became quite good at ingesting large amounts of data, planning out tasks, and taking action on behalf of users. (Read more about this progress in our special report, AI for Startup Leaders.)
With this leap your company's internal agents no longer need to be trapped inside one pre-defined workflow per team. The agent can be universal, while task-specific expertise can be defined in skills.
Adapt’s skills are "modular knowledge packages" that teach the agent how your team works. They can include domain knowledge, company-specific templates, style guides, past examples, step-by-step workflows, and your proprietary best practices. Adapt loads the relevant skill when the task calls for it, which enables the one universal agent to handle expert work without losing the broader company context.
What universal agents make possible
Work slop happens when people create AI output in isolation and drop it on teammates without the shared context needed to judge or improve it. A universal agent helps teams collaborate on ideas and tasks from the start. Teammates steer the agent's work together instead of cleaning up disconnected output after the fact.
Universal agents also reduce the pressure that creates shadow AI. Employees reach for rogue specialized agents when their team does not have access to the same AI capabilities other functions already have. A sanctioned universal agent gives teams the powerful intelligence they need, while keeping the work inside the company’s visibility and control.
Beyond quality and security, universal agents unlock the creative promise that made AI exciting in the first place. A universal agent enables the people closest to a problem to design new ways to operate.
An internal example: I vibe coded an internal dashboard that pulled from CRM, billing, product telemetry, and analytics, and displayed key metrics on a monitor everyone in the office could see. No designer was needed because it used our design system skill. No engineering or data integration work because Adapt just connects to the systems and pulls data. Zero barrier between initiative and execution - that's what a universal agent can deliver.
How AI native teams use Adapt
Stamped, an e-commerce reviews platform, connected 15 tools and data sources to Adapt, spanning Slack, GitHub, HubSpot, Notion, Zendesk, Gong, Linear, Mixpanel, GA4, Neon, Mantle, Store Leads, and Resend. A weekly Gong-and-Slack scan now collapses customer feedback into one report, and a non-engineer on the team built a Zendesk-analysis dashboard inside Adapt that the rest of the company actually uses.
"That's something Claude can't do," the operator said, and the reason is structural. The job needed action across several systems, not insight inside one.
Wander, a luxury travel and hospitality startup, started using Adapt as a data-access tool for one team, then later rolled it out as a company-wide operating layer.
More than 140 team members use Adapt daily with more than 590,000 messages sent in under seven months. Instead of waiting on a data engineer, teams can ask Adapt inside Slack and get answers from connected systems in seconds.
Adapt uses Adapt the same way internally. In addition to the CEO's dashboard, Engineering turns RFCs into Linear issues, marketing runs competitive intelligence and SEO monitoring from Slack, sales syncs pilot usage from BigQuery into HubSpot. The same universal agent operates across engineering, marketing, sales, product, and leadership without becoming five disconnected assistants.
FAQ
What is a universal AI agent for work?
A universal AI agent for work is one agent connected across a company’s apps, data, and workflows. It lets teams share context, take action across systems, and reuse the work the agent creates.
Why are specialized agents constrained?
Specialized agents are constrained because they are usually built around one product or one function. They can accelerate narrow work, but they cannot solve cross-functional problems.
Are universal agents smarter than specialized agents?
Universal agents like Adapt are smarter because they can discover context across more systems. They do not have to assume the answer lives inside one tool. They can follow the work wherever it leads to deliver a result.
What kind of teams benefit most from a universal agent?
Teams that depend on cross-functional work benefit most. Sales, marketing, support, product, engineering, and leadership all need answers that live across systems. A universal agent is useful when the work requires more than one app, one data source, or one department.
