이 블로그 글의 실습 코드는 어디서 받을 수 있나요?

각 글 하단에 첨부된 깃허브 링크에서 실습 코드를 내려받을 수 있습니다.

Why Anthropic’s $100 Million Training Program Includes Two Assessments

Anthropic’s Claude Frontier Academy requires engineers to pass a practical assessment and a reassessment after a 12-week workplace residency to earn its final credential. Here is what that structure can teach organizations about assessing employee AI skills, setting meaningful training completion standards, and measuring progress with Codepresso’s AI Fluent.

One-Minute
Summary 💡

On October 2, 2026, Anthropic announced Claude Frontier Academy, a $100 million initiative to train 10,000 Frontier Deployed Engineers (FDEs) by the end of 2027.

$100M investment 10,000 FDE target 12-week residency 2 assessments

The program includes two assessment milestones. Participants must first pass a simulated enterprise deployment assessment before entering the residency.

1. Practical assessment Complete a simulated enterprise deployment before entering the residency.

2. Workplace application Lead a real Claude project at the participant’s own company during the 12-week residency.

3. Reassessment Demonstrate capability again after the workplace project to earn the final credential.

Organizations can apply a similar structure to employee AI training: establish a baseline, assess performance through practical tasks, provide opportunities for workplace application, and reassess using consistent criteria.

You have added “96% completion rate” to your year-end training report. But does that number mean employees can now do their jobs more effectively with AI?

For many learning and development leaders, that is a difficult question to answer.

An AI training program developed by Anthropic offers a useful example of how to address it. Its $100 million initiative includes two assessments on the path to the final credential, connecting classroom learning with demonstrated workplace performance.

AI competency assessment evaluates both how employees use AI to solve realistic work tasks and the results they produce. By scoring performance against defined criteria, organizations can generate evidence to support training, role assignment, and hiring decisions.

This article examines Anthropic’s approach and explains how organizations can adapt its underlying structure to assess AI skills across their workforce.

What Is Claude Frontier Academy, and Why Does It Include Two Assessments?

Claude Frontier Academy is a training initiative announced by Anthropic on October 2, 2026, with a $100 million investment and a goal of developing 10,000 Frontier Deployed Engineers by the end of 2027.

Its first program, the Frontier Deployed Engineer Residency, follows a model inspired by medical training: participants learn from practicing professionals, work through realistic scenarios, and pass assessments before being recognized as ready to work independently.

The inaugural cohort includes engineers from Accenture, Bain, Capgemini, Commonwealth Bank of Australia (CBA), Deloitte, McKinsey, Morgan Stanley, and Novo Nordisk. The program operates in San Francisco, New York, and London.

Participation is by employer nomination. Each engineer enters with a specific Claude project they will lead after returning to their organization.

The program has four stages:

Stage	What Participants Do	Milestone
1. In-person intensive: several days	Work with Anthropic engineers and certified instructors on simulated enterprise deployments, covering use-case selection, security review, and handover	Become eligible for the practical assessment
2. Practical assessment	Complete a scored assessment using an unfamiliar scenario	Earn the Claude Resident Engineer badge
3. Twelve-week residency	Lead a real Claude project at their own company, supported by Anthropic engineers and peer learning within the cohort	Become eligible for reassessment
4. Reassessment	Complete another assessment at the end of the residency	Earn the Claude Frontier Deployed Engineer badge, with the first recipients expected in early 2027

Anthropic’s explanation points to the reason for this two-stage evaluation.

Steve Corfield, Anthropic’s head of global business development and partnerships, described how a small group of motivated people, with the right skills, access to Claude, and a deep understanding of their organization, can transform a business. The difficult part, he explained, is developing people with that depth of capability.

The assessments connect two different achievements: learning how to perform a task in a training environment and making it work inside an organization.

The scale of Anthropic’s existing training activity also provides context. Professionals from 46,000 companies in its partner network have already earned more than 175,000 Claude credentials, while nearly 4,000 participants have completed its immersive Basecamp program.

Codepresso sees the addition of workplace projects and reassessment at this advanced level as a sign that AI training is moving toward a stronger emphasis on demonstrated capability.

What Are the Main Ways to Assess Employee AI Skills?

Organizations commonly use five approaches: satisfaction surveys, completion records and certificates, multiple-choice knowledge tests, practical task assessments, and reassessment after workplace application.

The first three are relatively easy to administer, but they do not directly demonstrate whether employees can perform work with AI. The latter two require more effort to design and deliver, but can provide measurable evidence of capability and progress when used consistently.

Assessment Method	What It Measures	Best Suited For	Limitation
Satisfaction survey	Participants’ perceptions of the learning experience	Improving training delivery	Does not establish whether competency improved
Completion rate or certificate	Attendance and course completion	Tracking mandatory training	Confirms completion rather than practical ability
Multiple-choice knowledge test	Understanding of concepts and terminology	Assessing foundational knowledge at scale	Does not directly observe workplace performance
Practical task assessment	The process and results of completing a work task with AI	Setting completion standards and informing role assignment or hiring	Requires relevant tasks and clear scoring criteria
Reassessment after workplace application	Changes in performance measured against consistent criteria	Reporting training outcomes and planning further development	Requires at least two assessment points

Anthropic’s residency combines the final two approaches: a practical assessment followed by reassessment after a workplace project.

If you are already comparing assessment providers, Three Criteria for Choosing an AI Competency Assessment Service explains what to look for in how each service measures performance.

How Can Practical Assessments Become Part of AI Training Completion Requirements?

To build meaningful completion standards, structure the program around four elements: a baseline assessment, role-specific tasks, shared scoring criteria, and reassessment after workplace application.

Anthropic’s sequence, an assessment after the in-person intensive, followed by another after the 12-week residency, illustrates the value of evaluating performance at more than one stage. For internal employee training, adding a pre-training baseline also makes improvement easier to measure.

1. Establish a Baseline Before Training

Measure employees’ current competency before the program begins. Without a baseline, it is difficult to show how much their skills have changed.

The results can also help determine which employees need which learning pathways.

2. Use Practical Tasks Relevant to Each Role

Choose tasks that reflect employees’ responsibilities. Examples include developing a campaign plan for marketing, analyzing interview questions for HR, or organizing settlement records for finance.

This follows the same principle as asking residency participants to arrive with a specific workplace project: learning should connect to work they need to perform.

3. Apply Consistent Scoring Criteria Across the Organization

Evaluate more than the final deliverable. Assess what employees instructed AI to do, what they checked along the way, and how they corrected or refined the output.

Tasks may differ by role, but a shared evaluation framework is essential if the organization wants to compare results across departments.

4. Reassess After Employees Have Applied Their Learning

Allow employees to use AI in their actual work for several weeks, then reassess them against the same criteria.

Comparing this result with the baseline provides a more meaningful measure for training reports than completion rates alone.

In a KBS News 9 interview, Codepresso CEO Donghoon Lee expressed the principle clearly: training should produce a change in competency that can be observed before and after the program. Read more in Codepresso’s Coverage of the KBS News 9 Report.

This question is particularly relevant as AI training for non-technical employees expands. Of the four non-technical AI training initiatives discussed in our October 2 article, only one explicitly described a post-training assessment.

How Does Anthropic’s Approach Compare with Codepresso’s AI Fluent?

Both approaches emphasize assessing performance through realistic tasks and measuring capability again after practical application. Their audiences and evaluation scope differ.

Claude Frontier Academy develops experienced, employer-nominated software engineers into specialists in deploying Claude within enterprises.

AI Fluent assesses employees across technical and non-technical functions, including marketing, HR, business administration, customer service, R&D, and software development, using a shared scoring framework.

Dimension	Claude Frontier Academy: FDE Residency	Codepresso AI Fluent
Audience	Experienced software engineers nominated by their employers; LLM development experience required	Technical and non-technical employees across seven tracks: two common tracks, one non-development track, and four software development tracks
Assessment timing	Practical assessment after the in-person intensive; reassessment after the 12-week residency	Pre- and post-training assessments, with quarterly reassessment to track changes over time
What is evaluated	Simulated enterprise deployment performance and execution of a real company project	Prompt histories, intermediate work products—including plans, skill configurations, and source code—and final deliverables
Results	Claude Resident Engineer and Claude Frontier Deployed Engineer badges	Four competency levels—Entry, Beginner, Intermediate, and Professional—plus individual reports and organization-level data
Primary purpose	Developing deep expertise in a small group of key specialists	Mapping workforce AI competency, evaluating training effectiveness, and planning further development

For a company seeking to develop a few engineers into Claude deployment specialists, Anthropic’s model addresses a focused need.

For an organization that needs to understand AI capability across its workforce and report training outcomes to leadership, a common assessment framework across job functions serves a broader purpose.

Codepresso highlights two strengths of AI Fluent:

Consistent criteria across roles. Although assessment tasks differ, employees in marketing and software development are evaluated against a shared framework.
Visibility into the working process. Prompt histories and intermediate outputs help show how employees worked with AI, even when their final deliverables look similar.

For organizations that want to connect assessment with training, AXMOS’s AX Grow follows a related structure.

Designed and operated with Codepresso as a core participating company, the program combines AI competency assessment, practical training, and an internal builderthon based on real workplace tasks over approximately 16 business days. Deliverables include assessment reports, builderthon projects, and a roadmap for expanding adoption.

Can This Assessment Structure Work for Non-Technical Teams?

Yes. The tasks change, but the sequence remains useful: Baseline assessment → practical tasks → workplace application → reassessment

For example, AI Fluent’s AI Agent-Based Workflow Efficiency track is a common track open to employees across job functions. Its Prompt Engineering track evaluates the ability to guide AI responses toward business objectives without coding.

Roles that require deeper technical or implementation expertise need a different level of evaluation. For FDEs who work directly in client environments and carry an AI solution through to implementation, candidate selection and workplace validation are especially important. We explore this distinction in How to Develop FDE Talent. Anthropic’s employer-nomination requirement reflects a similarly selective approach.

Codepresso’s view is that organizations investing in AI training should report before-and-after competency scores measured against consistent criteria, alongside completion rates.

The evidence needed to evaluate a training program should be planned before the program begins. That gives learning leaders a clearer basis for explaining what improved, where gaps remain, and which training should come next.

Frequently Asked Questions

What Is Claude Frontier Academy?

Claude Frontier Academy is a training initiative announced by Anthropic on October 2, 2026, backed by a $100 million investment and a goal of training 10,000 Frontier Deployed Engineers by the end of 2027.

Its first offering, the FDE Residency, includes a practical assessment based on a simulated deployment and a reassessment after a 12-week workplace project. Participants earn credentials at these assessment milestones.

What Is an AI Competency Assessment?

An AI competency assessment evaluates how an employee uses AI to complete realistic work tasks, including both the working process and the final result.

Performance is scored against defined criteria to support training, role assignment, and hiring decisions.

Practical assessments that examine instructions to AI, intermediate checks, and final deliverables offer more direct evidence of workplace capability than knowledge tests alone.

Does an AI Training Certificate Prove an Employee Can Use AI Effectively?

A completion certificate confirms that an employee finished a course. Their ability to produce useful results at work needs to be established through a practical assessment.

Anthropic introduced its residency with practical assessment and reassessment requirements after more than 175,000 Claude credentials had already been earned across its partner network.

When Should Employees Be Reassessed After AI Training?

To understand how well learning transfers to the workplace, reassessment should follow a period of practical application.

Anthropic reassesses participants at the end of its 12-week residency.

Codepresso AI Fluent supports a pre-training baseline followed by quarterly assessments using consistent criteria to track progress.

How Can We Plan an AI Competency Assessment for Our Employees?

Contact Codepresso at sales@codepresso.kr with your organization size, the job functions you want to assess, and any AI training you currently offer or are planning.