March 2026
Defining AI-Native Work at Tribe
An executive summary of the AI Enablement Research synthesis
Jeremy Kirshbaum, Craig Barowsky, Andrew Enns
Six dimensions define what “amazing” AI-native work looks like at Tribe. Each is measured on five levels, with L4 anchored to named people doing real work and L5 aspirational — no one is there yet across all dimensions. The goal is to move the median level in each dimension.
Dimension 1 of 6
Context Architecture
How you structure your information environment — connections, documents, persistent contexts, and organizational knowledge — so AI can reason over your work without you re-explaining it every session.
Absent
Starts fresh each session. Copies and pastes context manually. No data connections. Hasn't yet thought of their work as decomposable into AI-legible steps. This is where most knowledge workers outside Tribe begin — the tools are new and the mental model of "structuring context for AI" doesn't exist yet.
Personal
Has Claude Projects or saved prompts. A few data connections. Can describe their workflow but hasn't externalized it for AI. Uses voice for context input. Session-to-session context is manual but intentional.
Systematic
Several data connections feeding real work data. Persistent contexts across sessions (CLAUDE.md files, project structures). Has decomposed their workflow into AI-delegable and human-judgment steps. Skills that encode recurring patterns.
Expert Exemplar
Many data connections orchestrated into coherent information flows. Designs context delivery systems, not just individual connections. Restructures documents and codebases for AI legibility. Skills and plugins that compose into larger workflows. Others adopt their context patterns.
Compounding
Designs and builds context systems that other people plug into — layered context architecture (org-wide defaults → project-specific → personal) that others adopt and extend without your involvement. Actively restructures shared knowledge (codebases, documentation, organizational knowledge bases) for agent legibility, not just personal legibility. Creates context templates and patterns that new team members inherit on day one. Your context architecture decisions become the default way others set up their environments.
Dimension 2 of 6
Verification Design
How you architect quality assurance into AI-assisted work — from gut-feel spot-checks to designed falsifiability with automated pipelines — rather than reviewing outputs after the fact.
Absent
Accepts AI output at face value or applies only gut feel. No structured review process. This is a natural starting point — when AI is new, you don't yet know what kinds of errors to look for or how often they occur.
Personal
Reads and edits AI output before using it. Spot-checks facts. Uses domain knowledge to catch obvious errors. Treats output as draft (32/37 do this). But verification is manual and inconsistent — depends on attention, not design.
Systematic
Has designed verification into their workflow: separate review sessions, different model for checking, structured evaluation criteria. Domain expertise provides natural verification mechanisms (finance: GL ties or doesn't; code: tests pass or don't). Catches errors through process, not just attention.
Expert Exemplar
Designs verification before starting work. Automated pipelines (Playwright for accessibility, test suites for code, cross-model adversarial checking). Objective functions that measure output quality. Verification infrastructure accelerates work rather than slowing it. Others adopt their verification patterns.
Compounding
Builds verification systems that others use — shared test suites, quality gates in deployment pipelines, automated standards enforcement that you designed and maintain. Your verification patterns become the way the team catches errors, not just the way you catch yours. Designs feedback loops where verification results improve future prompts and constraints. Tracks quality metrics over time and surfaces regressions. Incorporates security and prompt injection awareness into verification architecture. Others produce higher-quality output because your verification infrastructure exists, without needing to design their own.
Dimension 3 of 6
Effective Shipping
How you get from idea to finished, delivered, maintained output — whether that's a deployed tool, a closed deal, a shipped proposal, or a repeatable workflow. Includes crossing the 80% cliff and building things that survive contact with real stakeholders.
Absent
Uses AI for conversation, ideation, or drafting but hasn't produced standalone deliverables. May have ideas for tools, workflows, or processes but hasn't started building them. This is common — many knowledge workers use AI effectively for augmenting existing work without creating new things, and that's a valid use pattern at this stage.
Personal
Produces working outputs that stay on your machine or in your own workflow — skills, scripts, prototypes, draft proposals, personal processes. May have multiple projects at 80%. Nothing has reached its intended audience. The graveyard of nearly-complete work.
Systematic
Has crossed the finish line. At least one output is deployed, delivered, or in active use by its intended audience — a tool on Railway, a proposal that closed a deal, a workflow adopted by a team. Builds complete end-to-end solutions, not just prototypes. Iterates with real stakeholder feedback. Scope discipline — knows when to stop.
Expert Exemplar
Production-grade outputs that others depend on — maintained tools, repeatable deal processes, frameworks that teams reuse. Multiple shipped deliverables actively maintained. Builds with quality that scales — not just "it works" but "it works reliably and others can build on it." Treats delivery and maintenance as first-class concerns, not afterthoughts.
Compounding
Builds infrastructure that enables others to ship — deployment templates, proposal frameworks, starter kits, playbooks, or reference architectures that make it trivially easy for someone else to go from idea to delivered output. Designs maintenance into what you build so your work doesn't require you to survive. Creates systems and processes that others extend — your shipped work becomes the foundation others build on, not just a standalone deliverable. Others ship faster and more reliably because of infrastructure you personally created and maintain.
Dimension 4 of 6
Role Expansion
How you use AI to absorb work that previously required someone else's expertise — and whether you encode that absorbed expertise into guardrails, constraints, and infrastructure rather than just doing it yourself.
Absent
Works within traditional role boundaries. AI assists within the lane but doesn't expand what they do. No adjacent-role work. This is perfectly appropriate for many roles and contexts — not everyone needs to cross boundaries, and many people are early enough in AI adoption that role expansion hasn't been relevant yet. (Distinct from strategic restraint in Delegation Calibration — L1 here is "hasn't considered it," not "chose not to.")
Personal
Has used AI to do something outside their traditional role at least once — a PM writing code, a sales lead building a dashboard, a designer doing data analysis. But it's ad hoc, not systematic. Quality may be uneven because domain expertise isn't encoded.
Systematic
Regularly operates across role boundaries with AI. Has absorbed 1-2 adjacent capabilities and produces reliable output. Beginning to encode absorbed expertise into constraints — templates, checklists, design systems — rather than relying purely on AI + judgment.
Expert Exemplar
Operates across 3+ traditional role boundaries simultaneously. Has encoded absorbed expertise into guardrails and infrastructure that maintain quality. Others benefit from the encoded expertise (design systems, quality constraints, templates). The PM who removed the front-end engineer — but only because the guardrails make the quality reliable.
Compounding
Encodes the expertise of multiple absorbed roles into infrastructure that others inherit — you personally build the design system, the testing framework, the quality constraints, or the deployment guardrails that let someone else cross a role boundary they couldn't cross before. Actively mentors others through their own role expansion, not just modeling it. Designs team workflows that account for dissolved boundaries (who does what when anyone can do anything). Your guardrails and encoded expertise are the reason other people can safely expand into roles they're not trained for.
Dimension 5 of 6
Knowledge Compounding
How your learning, building, and sharing creates durable value beyond yourself — through artifacts others can fork, reference implementations others build on, and knowledge that compounds organizationally rather than staying trapped in your head.
Absent
Learns and builds independently. Hasn't yet started sharing work or discovering what others have built. Knowledge stays with the individual — not because of reluctance, but because sharing infrastructure and habits haven't formed yet. (The duplicate-building problem — Emile and Carlos built the same account tracker independently — is a symptom of this, not a personal failing.)
Personal
Shares when asked or when something feels finished. Engages with some shared content (reads #upskilling, attends power hours). But sharing is reactive, not proactive. Learns from pairing sessions but doesn't seek them out.
Systematic
Shares work proactively — posts in channels, presents at sessions, creates documentation. Actively seeks out what others have built. Learns from multiple sources. But knowledge transfer is still social (presentations, Slack posts) rather than artifact-based. Others are aware of their work but may not be building on it.
Expert Exemplar
Creates artifacts that others fork, extend, and build on without needing the original creator. Reference implementations that become standards. Others cite their work, use their tools, build on their patterns. Influence is measurable through cross-references and tool adoption — not just presentation attendance. Also a strong inbound learner — rapidly absorbs others' innovations.
Compounding
Builds the systems that make knowledge compound without relying on individual effort — you personally create the discoverable directory, the capture mechanism, the cross-project learning infrastructure. Designs artifacts specifically for forkability (documented, modular, with clear extension points) rather than just sharing what you built. Actively closes learning loops: when you discover something on one project, you encode it into a shared resource that other projects automatically benefit from. Builds the "bleeding edge → graduated" pipeline where others' work-in-progress is surfaced and matured. Your knowledge infrastructure is the reason the organization learns faster than the sum of its individuals.
Dimension 6 of 6
Delegation Calibration
How you judge what to give AI versus keep for yourself — including when NOT to automate — based on an accurate mental model of what AI can reliably handle in your specific domain and context.
Absent
Uses AI for everything without differentiation, or hasn't started using AI at all. No mental model yet of what AI handles well vs. poorly. This is a starting point — when AI is new, you don't yet have the experience base to calibrate.
Personal
Has a rough sense of what AI is good at in their domain. Delegates routine tasks, keeps "important" ones. But the boundary is based on feeling rather than evidence. May over-delegate in unfamiliar areas (the verification asymmetry: delegates most in exactly the domains where they can't check the output).
Systematic
Can articulate what AI handles well and poorly in their specific domain. Adjusts delegation based on stakes, domain difficulty, and verifiability. Has a mental model of AI capabilities that updates with experience. Knows when to preserve cognitive friction deliberately.
Expert Exemplar
Precise mental model of AI capabilities — knows exactly which tasks, which domains, which stakes warrant delegation vs. retention. Treats AI like "a good employee on day one" (needs onboarding, needs checking, has strengths and blind spots). Strategic restraint is a deliberate, articulated choice, not avoidance. Can explain their delegation philosophy and why it works for their context.
Compounding
Codifies your delegation philosophy into shareable guidelines, playbooks, or decision trees that help others calibrate without having to learn everything through trial and error. Actively updates these as AI capabilities change — you are the person who notices when a model upgrade makes a previously-unreliable delegation safe, or when a new risk emerges. Mentors others on domain-specific calibration (what works in finance vs. infrastructure vs. design). Designs security-aware delegation patterns and teaches others where AI delegation creates attack surfaces. Others make better delegation decisions because of frameworks you built and maintain.
Implications
Business Impact
The heuristic: each level increase on a single dimension adds roughly 12% to a person’s effective capacity. The dimensions compound. What each one contributes:
- Context Architecture— eliminates per-session rebuild time. One GM’s deal review system replaced what would be a research analyst function. Another’s orchestration system absorbed executive-assistant work.
- Verification Design— makes velocity safe. Automated verification absorbed a QA function and enabled removing a front-end engineer from a project.
- Effective Shipping— converts prototypes to delivered value. One SA produces client-facing prototypes that previously required an engineering team and weeks of calendar time.
- Role Expansion— collapses specialist bottlenecks. One PM operates across four role boundaries; the team removed a specialist role entirely.
- Knowledge Compounding— multiplies everyone else’s gains. One person’s forkable artifacts influence nine others’ practices without meetings or documentation drives.
- Delegation Calibration— prevents waste. Practitioners at L4 articulate exactly where AI fails in their domain, avoiding the slop that erodes client trust.
One level up across all 6 dimensions = roughly 2x per person.