What we hear in the field.
Raw observations from direct conversations with engineering leaders and decision-makers navigating AI adoption. No theory. No projections. What's actually happening.
16 observations
Filter by function
Field Observation: An Operations Analyst Wanted Easier Reporting. The Audit Found 22 Tasks Behind That Simple Ask.
Nuvepro Workforce Audit
A second audit at the same learning experience platform company, this time for Operations Analyst / BI Consumer (KPI Dashboards). The challenge was deceptively simple: be able to query reports with ease. The audit found a 9/77/14 split across 22 tasks.
Only two tasks were fully automatable: applying and adjusting dashboard filters for ad-hoc questions, and documenting frequently asked questions with step-by-step query runbooks. Both are repetitive, rule-based tasks that follow fixed patterns.
The 17 augment tasks span the full BI consumer workflow: identifying which dashboard contains the right metric, translating business questions into dashboard interactions, navigating between snapshot and trend views, validating metric definitions, writing natural-language queries in Gemini, interpreting outputs, and maintaining a KPI glossary. The audit also surfaced broader operations tasks: process improvement, workflow streamlining, inventory planning, KPI establishment, and policy formulation. All augmentable because AI can draft recommendations, but the analyst must validate against operational reality.
Three tasks were classified as human-only: managing business resources, overseeing day-to-day operations, and supervising staff. These are management responsibilities that require judgment about people and priorities, not data.
Gemini was the AI tool, notable because this is the first audit where a Google AI product appeared instead of ChatGPT or Microsoft Copilot. The team chose it specifically for natural-language querying of BI dashboards.
Key Observations
- A simple challenge ('query reports with ease') expanded to 22 tasks across BI navigation, metric validation, stakeholder triage, and operational management.
- Gemini was chosen specifically for natural-language querying of dashboards, the first audit showing a Google AI tool instead of ChatGPT or Copilot.
- The 9/77/14 split is consistent with other analyst roles: most tasks are augmentable because they need human interpretation, not just data retrieval.
- Three human-only tasks are all management responsibilities (resources, operations oversight, staff supervision), reinforcing that management stays human across every role we have audited.
- This is the second Degreed audit. Combined with the CSM audit (32/53/16), a picture emerges: their CS team has more automation potential than their ops team.
Relevance to Workforce Audit
This audit is valuable for two reasons. First, it shows how a simple user challenge ("query reports with ease") expands into a complex task landscape that touches BI navigation, data validation, stakeholder management, and operational oversight. The audit process surfaces this hidden complexity.
Second, comparing both Degreed audits reveals an organizational pattern: Customer Success (32% automate) has more automation potential than Operations (9% automate). This suggests that customer-facing operational tasks (data pulls, milestone monitoring) are more automatable than internal operational tasks (process improvement, workflow design), likely because internal operations require more contextual judgment.
Field Observation: A Finance Analyst Role With 25 Augmentable Tasks and Only 2 That Can Be Fully Automated
Nuvepro Workforce Audit
A workforce audit of Financial Analyst / Finance Operations Analyst at a global research and advisory firm found a 7/86/7 split across 29 tasks. The pattern reinforces what we saw in the NASCAR FP&A audit: finance is almost entirely an augment story.
Only two tasks were fully automatable: maintaining an analysis log with version control, and creating job description templates. Everything else needs a finance professional in the loop.
The 25 augment tasks span the full analyst workflow: gathering data from ERP and billing systems, cleaning and normalizing it, reconciling totals, building KPI calculations, performing variance and trend analysis, creating models, generating charts, writing narratives, responding to ad-hoc requests, and identifying anomalies. The audit also flagged several process automation opportunities: invoice processing, audit prep, month-end close, financial reporting distribution, approval workflows, due diligence, KYC checks, and contract redlining. All classified as augment because each requires human validation before the output can be trusted.
The two human-only tasks were financial domain knowledge (GAAP concepts, chart of accounts structure, cost allocation logic) and critiquing client strategies. These are judgment and expertise tasks that define senior analyst value.
ChatGPT was the AI tool in use. The challenge was straightforward: basic data analysis. But the audit revealed that "basic data analysis" in a finance context touches 29 distinct tasks across the full data-to-insight pipeline.
Key Observations
- Second finance audit in a row showing near-zero automation. The 7/86/7 split mirrors the NASCAR FP&A result (0/95/5), confirming finance is fundamentally an augment function.
- 29 tasks were identified from a challenge described as 'basic data analysis,' showing how audit depth reveals hidden complexity in seemingly simple roles.
- Process automation tasks (invoice processing, month-end close, reporting distribution) were classified as augment, not automate, because each requires human validation in a regulated context.
- The two human-only tasks are both about judgment: GAAP domain knowledge and strategic critique. These cannot be delegated to AI without professional risk.
- ChatGPT is the tool again. Two finance audits, two teams using general-purpose LLMs rather than specialized finance AI platforms.
Relevance to Workforce Audit
This is the second finance role audit showing the same pattern: AI augments nearly everything, automates almost nothing. Combined with the NASCAR FP&A audit (0/95/5), a benchmark is emerging for finance roles: expect 0-10% automate, 85-95% augment, 5-10% human-only.
The implication for organizations is clear: the ROI of AI in finance comes from speed and throughput, not headcount reduction. Finance teams using AI will close books faster, produce better narratives, and respond to ad-hoc requests more quickly. But they will still need the same number of analysts to validate, interpret, and make judgment calls.
Field Observation: HR Communications at a Financial Services Firm Is a Drafting Problem, Not an Automation Problem
Nuvepro Workforce Audit
A workforce audit of HR Communications / Benefits Communications Specialist at a global financial services firm found a 15/75/10 split across 20 tasks. The challenge was specific: use AI to draft communications to the company.
Three tasks were fully automatable: adapting announcements into multiple formats (email, intranet, FAQ, manager talking points), scheduling and publishing to the right channels, and maintaining version control and audit trails. These are mechanical distribution tasks that follow fixed rules.
The 15 augment tasks reveal why HR communications is harder than it looks. Drafting the initial announcement requires approved templates, tone guidelines, and plain-language clarity while preserving legal accuracy. Creating FAQs means knowing what HR can answer versus what goes to the vendor. Content validation requires checking against plan summaries and legal disclaimers. In a regulated financial services firm, every communication needs compliance language, regional localization, and stakeholder review cycles. AI drafts the first version, but a specialist shapes it for accuracy, tone, and compliance.
Microsoft Copilot was the tool of choice. The augment pattern here is classic copilot territory: AI generates the draft, the specialist edits for tone, accuracy, and regulatory requirements.
The two human-only tasks were career management support and leading/developing a team of HR Specialists. Both are relationship and mentorship tasks that require trust, not text generation.
Key Observations
- HR communications is 75% augment because drafting is the easy part. The hard part is regulatory accuracy, tone calibration, and stakeholder alignment.
- Only 3 of 20 tasks could be fully automated, all in the distribution and record-keeping category.
- In regulated financial services, even routine benefits announcements need compliance language, legal validation, and localization across regions.
- Microsoft Copilot fits this role well: draft generation, subject line variants, tone adjustment, and format adaptation are all within its capabilities.
- The two human-only tasks are both about people, not content: career guidance and team leadership.
Relevance to Workforce Audit
This audit shows that communications roles in regulated industries have higher augment percentages than you might expect. The drafting itself is augmentable, but the compliance review, stakeholder coordination, and regional localization keep a human in the loop for most tasks.
The 15/75/10 split provides a benchmark for HR communications and internal communications roles, particularly in financial services and other regulated industries. Organizations in less regulated sectors may see slightly higher automation percentages for distribution and formatting tasks.
Field Observation: Guest Relations at a Theme Park Giant Is 73% Augment, Not Automate
Nuvepro Workforce Audit
A workforce audit of Guest Relations / Customer Success Specialist at a major entertainment company revealed that nearly three-quarters of the role's tasks fall into the augment category. Only 12% can be fully automated, and 15% remain entirely human.
The challenge was improving customer-facing difficult scenarios. The audit broke the role into 26 tasks spanning complaint triage, resolution coordination, CRM documentation, follow-up communications, pattern identification, and cross-functional stakeholder work.
The three automatable tasks were documentation-heavy: CRM case logging, follow-up email dispatch, and maintaining digital resource libraries. Everything else requires a human in the loop because guest interactions are high-emotion, high-context, and happen in real time. AI tools like Microsoft Copilot can summarize conversations, propose severity ratings, generate intake checklists, and draft resolution options, but the specialist still confirms, overrides, and manages the interaction.
Four tasks were classified as fully human: handling escalations with highly upset guests, de-escalation in time-sensitive situations, outcome-focused customer guidance, and delivering brand-consistent guest experiences. These require judgment, empathy, and the ability to read a room that no AI model can replicate.
Key Observations
- Guest-facing roles are overwhelmingly augment (73%), not automate. AI helps with prep and documentation, but the human runs the interaction.
- Only 3 of 26 tasks could be fully automated, all in the documentation and follow-up category.
- De-escalation, conflict management, and brand-standard delivery remain entirely human. These are the tasks that define the role.
- The AI value is in speed, not replacement: faster triage, faster documentation, faster coordination with internal teams.
- Microsoft Copilot was the tool of choice, used across triage, intake, resolution generation, and case documentation.
Relevance to Workforce Audit
This audit confirms a pattern: customer-facing roles in high-touch industries are augment-dominant. Organizations auditing similar roles should expect minimal full-automation opportunities and should invest in training staff to use AI as a copilot during live interactions rather than trying to automate the interaction itself.
The 12/73/15 split provides a benchmark for other guest relations, hospitality, and customer experience roles. If your audit shows significantly higher automation percentages for a similar role, the tasks may be mis-classified or the role definition may be narrower than expected.
Field Observation: Freeing Up Customer Success Managers Means Automating the Ops, Not the Relationships
Nuvepro Workforce Audit
A workforce audit at a learning experience platform company asked a specific question: how do we free up Customer Success to spend more time talking to customers and less time on operational tasks?
The audit classified 19 tasks and found a 32/53/16 split. Six tasks were fully automatable, ten could be augmented with AI, and three must remain entirely human.
The automatable tasks were exactly the operational burden the team wanted to shed: pulling health and adoption data, cleaning and normalizing reports, preparing meeting briefs, creating and assigning internal tasks from meetings, monitoring onboarding milestones, and tracking usage data. These are the tasks that eat hours every week and produce no customer-facing value.
The augment tasks are where AI becomes a drafting partner: generating stakeholder reports, capturing and structuring meeting notes, converting notes into follow-up emails, updating CRM records, building onboarding sequences, and identifying expansion signals. The CSM still reviews, edits, and sends, but the first draft and data gathering happen automatically.
The three human-only tasks are relationship-core: ensuring customers realize long-term value, acting as the bridge between customer and company, and fostering personalized engagement. These are the tasks the team wanted more time for.
Key Observations
- The 32% automate slice maps exactly to the operational tasks the CS team wanted to eliminate: data pulls, report prep, task creation, milestone monitoring.
- AI tools (n8n + ChatGPT) were chosen for workflow automation and content generation, not a single enterprise suite.
- The three human-only tasks are all relationship tasks. The audit confirms that CS is fundamentally a relationship role, not an ops role.
- Augment tasks (53%) are about making the CSM faster at things they already do: meeting notes, CRM updates, onboarding templates, churn signals.
- The challenge framing mattered. Asking "free up time for customer conversations" led to a clean separation between automatable ops and human-only relationships.
Relevance to Workforce Audit
This audit demonstrates how challenge framing shapes the output. The team did not ask "how do we automate CS?" They asked "how do we free up time for what matters?" That led to a task breakdown that cleanly separated automatable operations from human-only relationships.
The 32/53/16 split is a useful benchmark for CSM roles at SaaS companies. The pattern is consistent: data gathering and report generation automate, drafting and CRM work augment, and relationship management stays human.
Field Observation: FP&A at a Sports Entertainment Company Is 95% Augment, 0% Automate
Nuvepro Workforce Audit
A workforce audit of Budget Management / FP&A at a major sports entertainment company produced one of the most lopsided splits we have seen: 0% automate, 95% augment, 5% human-only across 22 tasks.
Zero tasks were classified as fully automatable. Every financial planning task requires human judgment at some point in the loop. Budget input collection needs validation against business context. Variance analysis needs someone who understands why numbers moved, not just that they moved. Scenario planning requires assumption design that reflects business strategy, not just mathematical sensitivity.
The 95% augment classification means AI (ChatGPT was the tool) can accelerate nearly every task: normalizing data across ERP systems, building forecast models, generating variance narratives, running sensitivity analyses, preparing reporting packs, and drafting executive summaries. But a finance professional must review, validate, and sign off on every output.
The single human-only task was stakeholder management and business partnering: influencing budget owners, negotiating trade-offs, and communicating financial implications in terms non-finance leaders understand. This is the irreplaceable skill in FP&A.
The absence of any automatable tasks is notable. Even data collection and reconciliation, which feel like automation candidates, were classified as augment because the FP&A analyst needs to understand the data to catch anomalies that automated checks would miss.
Key Observations
- Zero automatable tasks in FP&A. Every task requires human judgment somewhere in the loop, even data reconciliation.
- 95% of tasks can be augmented with AI: faster modeling, automated first-draft narratives, quicker variance identification.
- The only fully human task is stakeholder management: negotiating budget trade-offs and translating financial data for non-finance leaders.
- ChatGPT is being used as the AI tool, suggesting FP&A teams are starting with general-purpose LLMs rather than specialized finance AI.
- Event-driven financials (race schedules, media rights, sponsorship) add complexity that makes full automation even less viable.
Relevance to Workforce Audit
This audit is a strong data point for finance and FP&A teams considering AI adoption. The 0/95/5 split suggests that AI in finance is almost entirely about speed and accuracy, not headcount reduction.
Organizations auditing FP&A roles should expect similar results: AI drafts the models, narratives, and reports, but the analyst validates everything. The ROI case for AI in FP&A is about throughput (more scenarios analyzed, faster close cycles, better narratives) rather than automation savings.
The event-driven nature of sports entertainment financials makes this an especially complex case. Simpler FP&A roles at companies with more predictable revenue may show slightly higher automation potential in data collection and reconciliation tasks.
Field Observation: The HR Team Started Building Agents Before Engineering Did
Nuvepro Field Interview
At a global payments technology company, HR Business Partners and People Ops Generalists have become some of the earliest non-technical agent builders. They built AI agents that work as natural-language policy coaches. Employees ask about leave policies, benefits, and manager actions, and get answers with links back to the actual source documents.
The HR team also built change-communications assistants that draft announcements, FAQs, and targeted message variants matched to internal comms patterns. A third use case came up on its own: talent workflow navigators that walk people through onboarding and case handling via SharePoint-based site agents.
What makes this worth noting is that these are not technical staff. They are HR generalists using low-code tools like Microsoft Copilot Studio and SharePoint site agents to build things that serve hundreds of employees. The agents took over the most repetitive slice of their work, answering the same policy questions over and over, while the HR partners focus on the parts that need judgment: employee relations, organizational design, and change management strategy.
Key Observations
- HR teams are building their own policy chatbots with no engineering support
- Three agent patterns came out of it: policy coach with source links, change-comms assistant for drafting announcements, and talent workflow navigator for onboarding
- Agents handle the repetitive stuff: the same policy questions asked hundreds of times
- HR partners shifted to the judgment-heavy work: employee relations, org design, change management
- The People Team, traditionally one of the least technical functions, was among the first to build production agents
- Low-code tooling (Copilot Studio, SharePoint agents) made it possible with no engineering dependency
Relevance to Workforce Audit
This observation challenges the assumption that agent building is an engineering function. The HR team's success shows that task-level classification (automate/augment/human-only) applies to non-technical functions just as powerfully:
1. **Automate**: Answering routine policy questions, generating standard communications variants, providing step-by-step process guidance. All moved to agents.
2. **Augment**: Drafting change communications (agent creates first draft, HR partner edits for tone and context), onboarding workflows (agent handles logistics, partner handles relationship building).
3. **Human-only**: Employee relations, organizational design decisions, sensitive change management conversations.
The Nuvepro framework applies directly: audit the HR team's tasks, classify them, build the agents for the automate tier, and train the people for the augment and human-only tiers. The fact that they did this with low-code tools means the training focus shifts from engineering skills to agent design, prompt craft, and governance.
Field Observation: The Comms Team Went from Brief to Full Campaign Calendar in One Sitting
Nuvepro Field Interview
Internal Communications and Events Coordinators at a global payments company are building AI agents that changed how campaigns get planned and executed. A Campaign Planner agent takes a brief and produces a calendar, channel mix, and targeted message variants with reusable checklists. Work that used to take days of manual coordination.
A second agent handles writing and visual production directly inside Microsoft 365: drafting leader emails, talking points, presentation slides, and event copy right where the work happens. Rather than switching between tools, the comms team stays in their existing M365 environment with AI assistance built into context.
The third pattern is a site agent that surfaces approved templates, brand assets, and prior campaign examples. Instead of digging through shared drives or asking colleagues where the latest brand guidelines live, the agent finds and delivers the right assets on demand.
The comms team said their bottleneck was never creativity. It was the operational overhead of coordinating channels, finding assets, and producing variants. The agents removed that coordination tax while keeping the creative judgment that makes internal communications work.
Key Observations
- Comms team takes a campaign brief and gets back a full calendar, channel mix, and message variants in one session
- Writing and visual production happens directly in M365, not by switching between tools
- A site agent finds approved templates, brand assets, and past campaign examples on demand
- The bottleneck was never creativity, it was coordination overhead. The agents removed that.
- Comms preserved their creative judgment while the repetitive production work got automated
- All agents built by non-technical comms staff using Copilot and SharePoint site agents
Relevance to Workforce Audit
The communications team's experience shows a pattern we see across non-technical functions: the highest-value work isn't what takes the most time. Creative judgment is the core skill, but it was buried under operational overhead.
1. **Automate**: Asset discovery, template retrieval, checklist generation, calendar coordination. Pure operational tasks that consumed the majority of time.
2. **Augment**: Campaign planning (agent generates structure, human refines strategy), message drafting (agent produces first drafts, human adjusts tone and positioning), variant creation (agent generates variations, human selects and customizes).
3. **Human-only**: Creative strategy, brand voice decisions, leader communication sensitivity, event experience design.
This maps directly to the Nuvepro framework: the task audit would reveal that 60-70% of comms team time goes to operational coordination (automate tier), with the remaining 30-40% being the strategic and creative work that the team was actually hired to do.
Field Observation: Meeting Prep That Took Hours Now Takes Minutes
Nuvepro Field Interview
Client Services Account Managers and Advisors at a global payments company built AI agents that changed how they prepare for client interactions. An Account Briefing Agent pulls together meeting prep from internal sources and public client news automatically. Work that used to mean hours of manual searching across multiple systems.
A Case Enrichment Agent delivers faster, better-cited answers to client inquiries by pulling from vetted knowledge sources. Instead of spending 30 minutes searching internal wikis and past case files, the agent surfaces relevant information with proper citations in seconds.
The third agent is a GTM Playbook Partner that drafts pitch materials and localized variants tied to existing go-to-market patterns. Account managers describe it as having a junior analyst who already knows every pitch the company has ever made and can adapt materials to any region or client segment.
The shift matters: account managers were spending 40-50% of their time on information gathering and material preparation. With agents handling that, they redirected time to relationship building, strategic account planning, and the consultative selling that actually drives revenue.
Key Observations
- Meeting prep that used to take hours of digging through CRM and email now takes minutes
- A Case Enrichment Agent gives faster, better-cited answers to client inquiries from vetted knowledge sources
- The GTM Playbook Partner drafts pitch materials and adapts them to any region or client segment
- Account managers were spending 40-50% of their time just gathering information and preparing materials
- That time now goes to relationship building, strategic planning, and consultative selling
- All built by non-technical client services staff using enterprise AI tools
Relevance to Workforce Audit
Client services is a high-value function where the human element (trust, relationship, strategic advice) is irreplaceable. But the preparation work that enables those conversations was consuming half the team's time.
1. **Automate**: Information gathering from multiple internal sources, public news monitoring, citation assembly, material localization for different regions.
2. **Augment**: Meeting prep synthesis (agent assembles the brief, human adds strategic context and relationship history), pitch material drafting (agent generates base materials, human customizes for the specific client relationship).
3. **Human-only**: Client relationship management, strategic account decisions, negotiation, trust-building conversations, exception handling.
The ROI case is straightforward: if account managers reclaim 40-50% of their time from information gathering, that time converts directly to more client-facing hours, deeper relationships, and higher revenue per account.
Field Observation: Product Ops Built a Gatekeeper That Checks Every Submission Before a Human Sees It
Nuvepro Field Interview
Product Operations teams at a global payments company built AI agents for two specific high-volume pain points. The first is an Intake Gatekeeper that does first-pass quality checks on incoming submissions: verifying completeness, checking for attachments, validating tags, and routing items to the right team. Before this agent, product ops staff manually reviewed every submission, and a significant chunk got sent back for missing information or wrong routing.
The second agent produces one-page status summaries and change narratives for leadership. These used to require an analyst to manually pull data from multiple tracking systems, stitch together status across workstreams, and write a narrative that executives could absorb in under two minutes. The agent keeps these summaries current, updating as underlying data changes.
What stands out is how precisely they picked their targets. Product ops didn't try to automate everything. They found the two tasks that ate the most time for the least strategic value. Intake QA is pure pattern matching that humans do reliably but slowly. Executive briefing is aggregation work where the value is in the format and narrative, not the data gathering. Both are textbook candidates for automation.
Key Observations
- Product ops built a gatekeeper agent that does first-pass QA on every submission before a human ever looks at it
- An executive briefing agent auto-generates one-page status summaries for leadership and keeps them current
- They picked their targets carefully: the two tasks that ate the most time for the least strategic value
- Intake QA is pure pattern matching. Humans do it reliably, but slowly.
- Executive briefing is aggregation work. The value is in the format and narrative, not the data gathering.
- A big chunk of submissions used to get sent back for missing info or wrong routing. The agent catches that upfront.
Relevance to Workforce Audit
Product ops demonstrates the ideal adoption pattern: find the highest-volume, lowest-judgment tasks and automate those first. This is exactly what the Nuvepro task audit recommends.
1. **Automate**: Submission completeness checks, attachment verification, tag validation, routing logic, data aggregation from tracking systems.
2. **Augment**: Executive briefing narrative (agent generates the summary, human adds strategic interpretation and recommendations), intake routing for edge cases (agent flags ambiguous items, human decides).
3. **Human-only**: Product strategy decisions, cross-team prioritization, stakeholder negotiations, change management.
The precision of their task selection is notable. They didn't try to automate the entire function. They ran a de facto task audit, found the two biggest time sinks, and built agents for exactly those.
Field Observation: Analysts Stopped Hunting Through Dashboards and Started Asking Questions in Plain English
Nuvepro Field Interview
Business Analysts and finance-adjacent analysts (non-technical) at a global payments company built AI agents that changed how they interact with data. A Natural Language Query companion lets analysts ask KPI and variance questions in plain language and get narrative answers. Not just charts, but explanations of what the numbers mean and why they changed.
The second pattern is what they call "Insight Blocks": curated, governed prompt and snippet libraries that standardize recurring analyses. Instead of each analyst writing their own queries and producing inconsistent outputs, the Insight Blocks provide a shared vocabulary of analytical patterns. An analyst can run a standard variance analysis, a quarter-over-quarter comparison, or a segment breakdown using pre-built, governance-approved templates.
The shift is from dashboard consumers to conversational analytics users. These analysts used to spend a big part of their day navigating Power BI dashboards, hunting for the right view, and then manually writing the narrative that goes with the numbers in a report. The NLQ agent collapses that: ask the question, get the answer with context, drop it into the report.
These are non-technical analysts. They don't write SQL or Python. Their skill is knowing which questions to ask and how to interpret the answers for business stakeholders. The agents handle the query execution and data retrieval. The analysts handle the interpretation and recommendation.
Key Observations
- Analysts stopped hunting through dashboards. They just ask questions in plain English and get narrative answers.
- Insight Blocks are curated, approved prompt libraries that standardize how recurring analyses get done
- The shift: from navigating dashboards to conversational analytics. Ask the question, get the answer with context.
- They used to spend hours finding the right dashboard view and manually writing the narrative around the numbers
- Non-technical analysts. No SQL, no Python. Their skill is knowing what to ask and what the answers mean.
- Governance is baked in: Insight Blocks make sure everyone uses consistent, approved analytical patterns
Relevance to Workforce Audit
The business analyst use case reveals a critical distinction in task classification: the analytical skill isn't in operating the tool. It's in knowing what to ask and what the answer means.
1. **Automate**: Data retrieval, query execution, dashboard navigation, standard variance calculations, quarter-over-quarter comparisons. All moved to agents via NLQ and Insight Blocks.
2. **Augment**: Narrative generation (agent produces data-backed narrative, analyst adds business context and caveats), anomaly investigation (agent surfaces unusual patterns, analyst determines if they're meaningful).
3. **Human-only**: Interpreting trends for business stakeholders, making recommendations, understanding organizational context that changes what numbers mean, presenting to executives.
The Insight Blocks pattern is particularly interesting. It's governance-first AI adoption. By curating and approving analytical patterns centrally, the organization ensures consistency while letting individual analysts work faster.
Field Observation: The Trainers Are Training Themselves First
Nuvepro Field Interview
Learning Facilitators, university partners, and field trainers at a global payments company are building AI agents that directly support AI adoption across the organization. A GenAI Adoption Coach provides scenario-based practice journeys that reinforce AI behaviors in the flow of work. Not separate training sessions, but contextual nudges and exercises tied to what employees are actually doing.
A Catalog and Logistics Concierge handles the operational side of learning: course discovery, session scheduling, materials Q&A. Employees ask the agent what training is available, when the next session runs, and what prerequisites they need. Questions that used to land in the L&D team's inbox as emails and Slack messages.
The third pattern is Live Demo Agent Patterns: workshop-ready examples drawn from internal agent-building sessions that trainers use to show what's possible. Instead of abstract slides about AI, trainers show working agents built by their own colleagues.
Here's the thing that stands out: the people responsible for teaching AI adoption are themselves the agent builders. They're not just explaining what agents do. They're building them, using them, and demonstrating them. That gives their training a credibility that no slide deck can match.
Key Observations
- The trainers are building agents before they teach anyone else. They lead by doing, not by slides.
- A GenAI Adoption Coach gives people scenario-based practice embedded in their actual workflow
- A concierge agent handles course discovery and scheduling so L&D isn't fielding the same questions all day
- Trainers demo working agents built by colleagues in the same organization. Real examples, not hypotheticals.
- Training credibility comes from practitioner experience, not presentations
- Adoption coaching happens in context during actual work, not in isolated training events
Relevance to Workforce Audit
The learning team's approach validates a core Nuvepro principle: the best AI training is hands-on, not theoretical. But they add another dimension: the trainers themselves becoming builders.
1. **Automate**: Course catalog Q&A, session logistics, materials distribution, scheduling. Operational L&D work moved to agents.
2. **Augment**: Adoption coaching (agent provides scenario-based exercises, facilitator personalizes for team context), demo preparation (agent patterns provide the base, facilitator adapts for audience).
3. **Human-only**: Facilitating live workshops, reading the room, adapting training in real-time, coaching individuals through adoption resistance, designing learning strategy.
The critical insight for the Nuvepro framework: AI adoption training must include agent building. The learning team proved that credibility comes from doing, not just teaching.
Field Observation: Managers Drowning in Information Built Agents That Turn It Into Decisions
Nuvepro Field Interview
People Managers at a global payments company built AI agents that address their core bottleneck: information overload. A Briefing and Decision Support agent takes meetings and documents and turns them into a structured format with options, risks, and follow-ups. Less time reading, more time deciding.
The second agent is an Adoption and OKR Nudger that prompts teams on AI usage habits, tracks progress against goals, and suggests next steps tied to OKR objectives. It acts as a persistent accountability partner, showing up when team adoption metrics drop or when goals need attention.
The management layer's adoption pattern looks different from individual contributors. Managers aren't building agents to do their work. They're building agents that help them lead their teams more effectively. The briefing agent doesn't make decisions; it structures information so the manager can make better decisions faster. The nudger doesn't manage the team; it keeps the manager aware of adoption patterns so they can step in at the right time.
This distinction matters: manager-tier agents are about decision quality and team enablement, not task execution. The managers said their biggest time sink wasn't any single task. It was the constant switching between information sources to stay on top of what their teams are doing.
Key Observations
- Managers were drowning in information. They built agents that turn meetings and docs into options, risks, and follow-ups.
- An OKR Nudger keeps tabs on team AI adoption and flags when goals need attention
- Manager-tier agents are about decision quality and team enablement, not task execution
- The biggest time sink for managers: constantly switching between sources to stay aware of what their teams are doing
- The agents don't make decisions. They structure information so managers can decide better and faster.
- The nudger acts as a persistent accountability partner for team adoption metrics
Relevance to Workforce Audit
The people manager observation reveals a separate tier of AI adoption that most frameworks miss: the management layer. Nuvepro's task audit classifies IC tasks, but managers have a different task profile.
1. **Automate**: Information aggregation from multiple sources, meeting synthesis, progress tracking, status report generation.
2. **Augment**: Decision structuring (agent presents options/risks/follow-ups, manager applies judgment), adoption monitoring (agent surfaces metrics, manager decides when and how to intervene).
3. **Human-only**: Team leadership, coaching conversations, performance feedback, organizational politics, building trust, navigating ambiguity.
The key insight: manager adoption of AI requires a different Skill Bundle than IC adoption. Managers need to learn agent supervision, delegation to AI, and how to use AI-structured information for better decisions.
Field Observation: Sales Teams Turning Meeting Notes Into Finished Proposals in One Session
Nuvepro Field Interview
Sales and Partnerships teams in B2B travel-adjacent contexts at a global payments company built AI agents that compress the sales preparation cycle. An Opportunity Packager converts meeting notes into structured proposals, localized collateral, and talk tracks. Work that used to require a sales ops analyst or several hours of a senior seller's time.
A Client Meeting Prep agent rolls up the latest product updates, client signals from CRM and internal systems, and prior materials into one comprehensive brief. Before this agent, sellers described their pre-meeting routine as a scavenger hunt across email threads, CRM notes, product wikis, and shared drives.
The team's adoption pattern is notable for its pragmatism. They didn't start with an ambitious AI transformation vision. They started with a simple question: what takes the most time before every client meeting? The answer led directly to the agents they built. No strategy deck, no steering committee, no multi-quarter roadmap. Just a team solving their own problem.
The risk they flagged: inconsistency. Different sellers building different agents with different prompt patterns, producing materials of varying quality and brand alignment. They saw the need for shared templates and governance but hadn't formalized it yet.
Key Observations
- Sales teams go from meeting notes to finished proposals in one agent session
- A meeting prep agent rolls up product updates, CRM signals, and prior materials into one brief
- Before this, pre-meeting prep was a scavenger hunt across email, CRM, wikis, and shared drives
- They started with a simple question: what takes the most time before every client meeting?
- No strategy deck, no steering committee. Just a team solving their own problem.
- Risk they spotted: inconsistency. Different sellers, different agents, different quality levels.
- They know they need shared templates and governance but haven't formalized it yet
Relevance to Workforce Audit
The sales team's experience illustrates both the promise and the risk of bottom-up AI adoption without organizational framework.
1. **Automate**: Information aggregation from CRM/email/wikis, product update compilation, prior materials retrieval, basic proposal structure generation.
2. **Augment**: Proposal customization (agent generates base, seller personalizes for relationship), talk track creation (agent drafts from patterns, seller adjusts for audience), collateral localization (agent adapts, seller validates).
3. **Human-only**: Relationship strategy, negotiation, pricing decisions, contract terms, trust building, reading client signals in conversation.
The governance gap they identified is critical. It's the gap between the Explorer stage and a mature AI-First organization. Individual agents work. But without shared templates, prompt governance, and quality standards, you get inconsistent outputs that can damage client relationships.
Field Observation: People Across Departments Started Building the Same Agent Without Talking to Each Other
Nuvepro Field Interview
A global payments company found a cross-functional cohort of non-technical Agent Builder starters spread across multiple departments. These are individuals who, regardless of their function, started building the same two agent patterns independently.
The first is a Team Knowledge Agent: a site-scoped FAQ and how-to resource for core processes and team assets. Every team's version is different, but the pattern is identical. Take the team's documented processes, policies, and frequently asked questions and make them searchable through a conversational interface. Instead of Slack-messaging a colleague asking "how do we do X?", team members ask the agent.
The second pattern is a Personal Productivity Agent: a lightweight action assistant for email and meeting synthesis, task drafting, and file discovery. These are personal tools, not shared resources, and they mirror each builder's individual work habits.
The organizational challenge is real. When dozens of employees across different functions independently build similar agents, you get duplication (multiple teams solving the same problem differently), inconsistency (no shared quality standards), and governance gaps (agents accessing data without formal approval). The company saw that this organic adoption needed structure. Not to shut it down, but to channel it.
This is the moment where individual AI adoption needs to become organizational AI strategy. The builders are already building. The question is whether the organization provides the framework, governance, and shared patterns to make it effective and safe.
Key Observations
- People across departments started building the same kind of agent without talking to each other
- Team Knowledge Agent: every team built their own version of a conversational FAQ for processes and policies
- Personal Productivity Agent: lightweight assistants for email, meeting notes, task drafting, file search
- The result: duplication, inconsistency, and governance gaps. Same problem solved different ways with no quality standards.
- Organic adoption needs structure. Not to stop the building, but to channel it.
- This is the tipping point: individual AI adoption has to become organizational AI strategy
- The builders are already building. The question is whether the organization gives them a framework.
Relevance to Workforce Audit
This observation captures the exact inflection point the Nuvepro framework is designed for. The organization has moved past Explorer. People are building. But it hasn't reached AI-First maturity where building happens within a governed framework.
1. **The duplication problem**: Multiple teams building team knowledge agents independently. A task audit would identify this as a shared pattern and recommend a template-based approach: one agent architecture, customized content per team.
2. **The governance problem**: Personal productivity agents accessing email, calendar, and files without formal data access approval. The audit would flag this and recommend a governance layer before scaling.
3. **The quality problem**: No standards for agent output quality. Some agents give great answers; others hallucinate or return outdated information. The audit would recommend evaluation frameworks and shared prompt patterns.
4. **The scaling opportunity**: The cross-functional cohort proves demand exists. The organization doesn't need to convince people to adopt AI. They need to give them the right tools, templates, and guardrails to do it well.
This is the strongest validation of the Nuvepro model: the workforce is already adopting AI. They need the framework to do it at organizational scale.
Field Observation: A 28-Year Engineering Veteran Who Can't Make the Business Case for AI
Nuvepro Field Interview
An Engineering Lead with 28 years of experience runs the networking division of a product company that builds networking equipment (hardware and software) for Managed Service Providers. He sees his team members picking up AI tools on their own. A manual tester started using Claude to write automation tests. A few staff members use Claude to knock out reports and documentation faster. A lead engineer uses GitHub Copilot for unit test cases.
The problem is, he still can't make a solid business case to roll Claude Code out to the whole team. He has anecdotes, not data. He can point to individual wins, but he has no systematic view of which tasks across his division should be automated, which need AI assistance, and which should stay fully human. Without that structured breakdown, he can't put a number on the ROI or give leadership a compelling reason to invest.
This is a textbook "AI Explorer" situation. Adoption is happening bottom-up, tool by tool, person by person. But there is no organizational framework to measure the impact or figure out the next step. He needs to move from scattered success stories to a department-wide task audit that maps every role to automate, augment, or human-only categories, puts numbers on the time savings, and lays out a phased rollout plan.
Key Observations
- A 28-year engineering vet sees his team using AI individually but can't build the ROI case for rolling it out to everyone
- Individual AI adoption is already happening: testers using Claude, engineers using Copilot, staff generating reports faster
- He can't get organization-wide buy-in because he has anecdotes, not a systematic view of which tasks should be automated
- No structured breakdown means no way to quantify ROI or present a convincing case to leadership
- Classic AI Explorer stage: bottom-up adoption with no organizational roadmap or framework
- He knows AI works. He just can't prove it at the division level without a proper task audit.
- The gap between individual wins and organizational ROI is the bottleneck, not the technology
- What he needs: a full task-level breakdown across roles, quantified savings, and a phased rollout plan
Relevance to Workforce Audit
This validates Nuvepro's core value proposition directly. The Engineering Lead's situation maps to the "AI Explorer moving up" segment:
1. **Current state matches Explorer profile**: Individual tool adoption (Claude, Copilot) without organizational strategy. Each person picks their own tool, uses it for their own tasks, with no cross-team visibility.
2. **The ROI gap is real**: He can't justify org-wide rollout because he lacks the task-level breakdown. This is exactly what the Workforce Analyzer solves: audit every role's tasks, classify into automate/augment/human-only, and quantify time savings.
3. **Specific automate candidates identified**: Writing automation tests from manual test cases (automate), generating reports and documentation (automate/augment), creating unit test cases (augment).
4. **The leadership gap**: As an engineering lead, he needs the organizational view. Which departments, which roles, which tasks, what tools, what ROI. Without this, individual wins stay individual.
5. **Nuvepro engagement path**: Run the Workforce Analyzer on his division, get the task breakdown, show ROI per role, build the rollout roadmap, justify the Claude Code investment to leadership.
This observation proves that even in tech-savvy teams with decades of engineering leadership, the gap between individual AI adoption and organizational AI strategy is the critical bottleneck.
From Our Audits
Aggregated insights from 40 completed workforce audits. Only roles with 3 or more audits are shown for statistical significance.
Software Engineer
EngineeringBased on 7 audits
Financial Analyst
FinanceBased on 3 audits
How We Gather These
Direct conversations with leaders navigating AI adoption in their organizations.
Direct Conversations
We talk to engineering leaders, department heads, and CXOs about what's actually happening with AI in their teams.
Generalized Specifics
Company names and identifying details are abstracted. Industries, roles, and patterns are preserved.
Real Patterns
Every observation maps to a pattern we see repeatedly: the gap between individual AI adoption and organizational strategy.