Leveraging Skills & The Frontier Firm
One of the fallacies I see with ambitious people that are early or mid-career is their desire to learn how to do what someone else does. They think that acquiring skills is critical to their success or ascension within a team.
However, it’s more important to learn how to leverage the skills of others. It enables greater output. It produces better outcomes. Don’t become a skills hoarder.
That same idea shows up in how two platforms are evolving what they offer.
Salesforce and HubSpot both had conferences earlier this month, Dreamforce and Unbound respectively, where their leadership and product teams talked about what they’re seeing with their customers.
Salesforce sees uneven adoption of AI within their customer base. To address that, they’re offering three front doors for users: Claudeforce, Slackforce, and Lightning (their existing UI). All of these are powered by AIforce.
Putting forced branding to the side, this distills to Salesforce offering a set of skills available to Claude and prebuilt agents on their core platform. They are also offering functionality for agents to perform actions in browsers to control websites and other SaaS apps, and released their own reasoning model based on one of Nvidia’s models.
Salesforce is also trying to monetize how agents interact with their platform, which means they’re likely seeing a world where not every employee needs a license to their products or they’re opportunistically trying to capture some additional revenue. They have not provided any specifics on how they intend to discern whether activity is from a person or an agent, but they’re prepared to bill you for both.
HubSpot’s CEO Yamini Rangan is taking an approach that centers around what data is available to AI with their Context Hub. She presented statistics during the keynote that performance across marketing, sales, and service use cases that leverage AI with bad context is worse than not using AI at all. When swapping that with good context, the performance with AI across those domains is significantly better than not using AI.
We could debate what “good” and “bad” mean as it relates to context for AI, or just take it at face value that better instructions and supporting data is intuitively going to be better for AI or people doing a task. HubSpot also sees AI as democratizing what can be built, and she goes on to say that “human intelligence has never mattered more” in the context of discerning where to focus effort and what to build.
While both of these offerings sound compelling, they’re also creating an additional degree of vendor lock-in while agentic harnesses continue to evolve.
I find it more compelling to have portability and modularity with skills, models, and a vendor-agnostic data source at this stage of the adoption curve. Mastering good skills, prompts, and secure access to data beats locking those into one front door. Avoid trying to master every platform’s set of AI features if you don’t have a solid core.
It’s also an appropriate time to start figuring out what seemingly manual work your teams are doing and how to begin transitioning it to automation with AI. Use the metaphor from last week’s newsletter: approach this like modeling with clay, not going straight to chiseling marble. If you want help working through how to do this, reach out and we can talk through how to approach it.
Let’s go a bit deeper on how applying AI can have positive and negative effects on psychological safety.
Jayshree Seth and Amy Edmondson offer four patterns to watch out for, which are detailed neatly in this HBR article. They demonstrate how AI can be used by employees to share and refine their perspective based on how safe and tolerant the organization’s culture is.
- AI either acts as a workaround for those that fear retribution;
- An equalizer when safety is not evenly developed within the culture;
- A sandbox to refine a position before sharing it;
- Or, as a tool leaders can use for honest feedback when they might not be receiving complete and comprehensive inputs from their subordinates.
In all of those cases, AI can positively affect an organization and its employees depending on what type of culture exists. This also has a low barrier to adoption: it’s simple to open your chat-based AI of choice, type or talk about a given scenario, and leave with a different perspective having gone through that process.
On the other hand, applying AI to track employees is probably not a great idea.
Friebel names two costs that arise from this practice in The Hidden Costs of Monitoring Employees with AI: installation and upkeep, and behavioral change. He also mentions a lot of common sense points like seeing blowback when experienced employees are required to use quality control checklists, and the negative effects that Amazon saw when using a leaderboard to see how many AI tokens its employees were using.
Absent is the question about why leadership wants to track employees though.
Microsoft identified this as “productivity paranoia” back in 2022, which initially started as an attempt to identify whether employees were still producing outputs as a result of remote and hybrid working arrangements due to Covid (side note: outcomes are way more important than outputs).
That gave rise to IT teams rolling out software that could track how and what employees were doing, mostly to understand utilization. Northeastern produced a report in May that flags how those employee monitoring vendors end up sharing data with third parties, often not disclosing that they will; some of that data even includes the precise location of employees.
Harvard Business School wrote in April that the surveillance economy has expanded to focus on how the work is being done to train models or automate based on employee actions. This has even expanded to the real-world with companies like Claru and Micro1 that collect video data to train models how to perform physical tasks.
This is practically the same method that Tesla uses to offer the best self-driving feature available. With a large collection of examples on what to do (and what not to do), AI can learn to automate the tasks that knowledge workers perform.
Microsoft sees the adoption of AI playing out as “the frontier firm”, whereby AI start acting as assistants, then become “digital colleagues”, and finally ascend to running entire business processes. The result is a workforce that is always-on and won’t be encumbered by days full of back-to-back meetings and distractions that arise every two minutes.
The frontier firm concept applies to existing companies and new entrants into the market as well. Your peers are looking at how to adopt AI to gain an edge and disruptors are also looking at how to apply AI within industries that have been late to adopting technology to streamline operations.