SUMM

Many companies now declare themselves "AI first," often expecting AI solutions before hiring people.
Big tech companies report increasing amounts of code written by AI, sparking debates about the future of software engineering roles.
Three main themes are introduced: the anatomy of an AI team, the evolution of generalists, and hiring strategies.

Companies are categorized as technology companies, verticalized/services companies, or tech-enabled companies.
Each type faces different challenges: tech companies may lack domain knowledge, services companies may face extremes in outcomes, and tech-enabled companies often struggle with technology implementation.
Technology adoption lags behind availability; factors like legacy processes and industry inertia matter more than technical limitations.

The key question is not whether technology is a limitation, but how it is applied within the team's structure.
Hiring AI researchers is only necessary at certain stages or for specific needs; many companies can achieve value without deep specialization.
The team’s primary activities include defining use cases, integrating with existing products, measuring ROI, finding data, testing workflows, building interfaces, and selling the solution.
A successful AI product requires cross-functional collaboration; no single role or specialty covers all needs.

It's essential to identify an organization’s real bottlenecks (e.g., shipping features, acquiring users, scalability, reliability).
Hiring should be prioritized to address these primary obstacles, with team structure adapting accordingly.

Generalists are valuable, especially in early stages or constrained budgets, due to their adaptability and broad skill sets.
Example from 2021: Built a team emphasizing model training, serving, and business acumen, focusing on practical skills rather than cutting-edge research.
As open source tools matured, newer teams leveraged existing platforms rather than building from scratch, freeing up resources for domain-specific needs.
The skill distribution within teams needs deliberate balancing based on current business priorities and available resources.

Existing teams can be optimized by distinguishing between inner loops (core daily tasks) and outer loops (strategic differentiation).
Key team nucleus includes model training, prompting, product requirements, model serving, domain expertise, and business case building.
Weakness in technical or domain areas will hinder product execution or market fit.

Generalists are essential during the initiation and transformation phases, providing adaptability and breadth.
As teams approach optimal performance, specialists may be necessary to achieve further incremental gains.

Team members must continuously build new skills, become domain experts, and be comfortable engaging with customers.
Functional prototyping, active participation by domain experts, and direct engineer-customer interaction are emphasized for faster feedback and better product-market fit.
Continuous learning should be structured into weekly cadences, focusing on both technical and business priorities.

Hire to maintain context or to execute on context; having too few team members leads to dropped priorities and poor execution.
AI has limitations in independently verifying or providing the expertise required for complex context management.
Human accountability remains crucial; systems need oversight.
Hiring should be guided by actual team composition needs and not by current hiring trends or hype.
Junior engineers and new graduates still hold value; trends dismissing entry-level positions should be questioned.

Interview questions must be relevant to actual job tasks, avoiding tests like generic coding challenges that can be solved by language models.
Collaborating cross-functionally and building teams based on domain requirements leads to better outcomes than simply hiring top technical researchers.
Ongoing learning and adaptation are necessary due to the fast pace of change in AI and technology.

Start team design with a clear understanding of what kind of team is needed for success.
Cross-functional teams will increasingly be the norm, with more overlap among roles.
Continuous learning should be a core part of team culture.
The pace of change in AI necessitates ongoing skill development and adaptability.

Structuring a modern AI team — Denys Linkov, Wisedocs