The Future of AI and Staying Current

Module 13 · Where AI Is Heading and How to Keep Up

Content current as of April 2026. AI trends evolve rapidly — the frameworks for evaluating them remain applicable.

Learning Objectives

Builds on: Modules 1-12 (the full course foundation)

By the end of this module, you will be able to:

  1. Describe the current state of five AI trends: agents, multimodal, small models, regulation, and embedded AI
  2. Apply the “hype vs. substance” framework to any new AI claim you encounter
  3. Explain the difference between augmentation and automation using real usage data, not speculation
  4. Maintain a 30-minute weekly routine that keeps your AI literacy current after this course ends
  5. Recognize when to ignore AI news entirely (more often than you’d think)

The Only Prediction That’s Always Right

Every prediction about AI’s future is wrong. The 2015 prediction that radiologists would be obsolete by 2020 was wrong. The 2022 prediction that ChatGPT was a passing fad was wrong. The 2023 prediction that image-generation would displace illustrators within a year was wrong. The 2024 prediction that AGI was imminent was wrong. The 2025 prediction that 95% of enterprise AI pilots would succeed was — documented by MIT — wrong by a factor of roughly 20.

So this module isn’t about prediction. It’s about learning to read current signals well enough that when the next prediction fails, you’re the one who saw it coming.

The test of AI literacy isn’t whether you know what will happen next. It’s whether you can tell the difference between a hype cycle and a structural shift while they’re both happening in front of you.

Common Misconceptions About AI’s Future
  • “AGI is around the corner” — Major labs describe their work as “paths toward AGI,” but there is no scientific consensus on whether current architectures can get there, and no agreed-upon test for when it would be achieved.
  • “AI will replace most jobs within 5 years” — The WEF projects 78 million net new jobs by 2030. AI changes roles; it rarely eliminates them entirely. The pattern is augmentation, not replacement.
  • “The best strategy is to wait and see” — By the time a technology stabilizes, the early adopters have already captured the premium. The EU AI Act’s August 2026 deadline isn’t waiting for anyone.
  • “If I read enough AI news, I’ll stay current” — Most AI news is noise. A 30-minute weekly routine with curated sources beats 5 hours of scrolling Twitter/X. Signal over volume.
  • “Open-source models will always lag behind proprietary ones” — DeepSeek-R1 and Llama 4 have closed the gap dramatically. The three-tier landscape (frontier/mid-tier/edge) is more useful than the open-vs-closed binary.

Trend 1 — Agents: no longer near-term, now production infrastructure

When this course launched in 2024, “AI agents” was a marketing term attached to demos. As of April 2026, agents are production infrastructure that companies pay for.

What changed in the last 12 months

  • OpenAI Operator (early 2025) was the first mainstream agentic consumer product. It’s now integrated into Codex with specialized GPT-5.4 Mini and Nano submodels for multi-model agent architecture.
  • Anthropic Claude Managed Agents launched in public beta April 8, 2026 — twelve days before this module was last updated. Production customers at launch: Notion, Rakuten, Asana. Already using underlying agent infrastructure: Netflix, Spotify, KPMG, L’Oréal, Salesforce. Pricing: Claude model usage plus $0.08/agent runtime hour.
  • Claude Code passed $2.5B annualized run-rate revenue by early 2026, doubling since the start of 2026.
  • Anthropic’s total annualized revenue: $14B as of February 2026, up more than 10× in three years.
  • 91% of enterprises use AI coding agents in production (Arcade.dev State of AI Agents 2026).
  • Microsoft Copilot Studio leads enterprise agent orchestration at 38.6% market share (VentureBeat, Q1 2026), with OpenAI at 25.7%.

The actual bottleneck in 2026

Twelve months ago, the question was “can agents do this task?” In 2026, the question is “can we plug this agent into our systems?”

46% of enterprises cite integration with existing systems — CRMs, ticketing tools, internal APIs, data platforms — as their primary challenge, not model capability (State of AI Agents 2026). The models are ready. The plumbing isn’t.

What this means for your career

The skills that matter for agent-era work are not “prompt engineering” narrowly. They are:

  1. Tool design — knowing which tasks to hand to agents vs. which to keep human-in-the-loop.
  2. Integration thinking — understanding data flows between systems.
  3. Observability — watching what agents actually do, not what they claim to have done.
  4. Governance — defining what agents are and aren’t allowed to do.
  5. Evaluation — deciding whether an agent’s output is good enough before it ships.

If you want one capability that will still be valuable five years from now, it’s evaluation. Models change every six months. Evaluation skill compounds.

The vendor lock-in question you can’t ignore

Both Anthropic’s Claude Managed Agents and Microsoft’s Copilot Studio are “platforms” that make agents easier to deploy at the cost of locking you into one vendor’s ecosystem. OpenClaw, an open-source agent framework launched in November 2025, hit 100,000+ GitHub stars by January 2026 — before Anthropic restricted its subscription access in April 2026, affecting 135,000+ running instances.

This tension — managed convenience vs. open-source control — is the single most important architectural decision enterprises are making in 2026. There is no universally correct answer. The right answer depends on your organization’s risk tolerance, skills, and regulatory environment. What is clear: lock-in risk is now a standard interview question for technical roles.

Trend 2 — Multimodal: text alone is no longer enough

A year ago, most AI products were text-in/text-out. In 2026, the leading models routinely handle text, images, audio, video, code, and PDFs in a single request. The architectural pattern has shifted.

The current state (April 2026)

Modality Leading models Current capability
Text + long context Claude Opus 4.7 (1M+ tokens), GPT-5.4 Pro (1M+), Gemini 3.1 Pro (1M+ multimodal) Entire books, codebases, meeting transcripts — in one context
Image generation Nano Banana Pro (Google), GPT Image 1.5 (OpenAI), Midjourney, Flux 2.0, Ideogram Near-photographic quality, character consistency across multiple images
Image input/analysis Claude (built-in), Gemini, GPT-5.4 Reads charts, diagrams, screenshots, handwriting. No separate OCR needed.
Video generation Veo 3.1 (Google), Kling 3.0 (Kuaishou), Seedance 2.0 (ByteDance), Runway Gen-4.5 Native 4K, up to 25 seconds per clip, synchronized audio, character consistency
Speech / audio OpenAI Advanced Voice, Google Gemini Live, Apple Live Translation Near-real-time translation, voice cloning (with safeguards), conversational flow

What happened to Sora?

OpenAI’s Sora launched as a standalone app in late 2025 and briefly topped the App Store. In March 2026, OpenAI shut down the Sora app and API, citing unsustainable compute costs, and folded the technology back into ChatGPT integrations. The episode is a case study in AI hype-cycle dynamics: extreme attention, unsustainable unit economics, sudden retreat. Veo 3.1 (Google, January 2026) is now the leading standalone text-to-video model for general use, with Kling 3.0 and Seedance 2.0 competing aggressively from China.

Why this matters for non-specialists

If your field touches any of the following, multimodal AI is already useful to you:

  • Marketing and communications: video generation has cut production time by 70% for many teams; AI-generated or AI-assisted video is projected to be ~75% of marketing video content by end of 2026.
  • Education: slide decks, worksheet diagrams, video explainers can now be generated from a single well-structured prompt (with verification).
  • Research: Gemini 3.1 Pro reads PDFs, charts, and code repositories in a single query. For literature reviews, the effective workflow is now “feed 30 papers in, ask structured questions out.”
  • Healthcare and science: multimodal models can analyze image + text together — medical image plus patient history, chart plus metadata.
The multimodal verification problem is harder

Text hallucinations are bad. Visual and audio hallucinations are worse — because they’re more convincing. A fabricated quote is easier to spot than a fabricated video. The verification skills you learned in Module 7 (fact-checking AI outputs) apply more strongly to multimodal content, not less.

The rule: if a video or image could plausibly be AI-generated, treat it as AI-generated until you have evidence otherwise. This is now a media literacy skill, not just an AI skill.

Trend 3 — Small models: specific, cheap, and on your phone

In 2024, if you wanted useful AI, you ran a frontier model via API. In 2026, that’s one of three deployment options — and often not the best one.

The three-tier landscape

Tier Parameter range Deployment 2026 examples
On-device (phones, IoT) Under 4B Smartphones, edge devices Gemma 4 E2B/E4B, Phi-4-mini (3.8B), Qwen 3.5 0.8B, Apple Foundation Model (~3B), Gemini Nano
Edge servers / laptops 4B–9B Local workstations, small servers Qwen 3.5 9B, Phi-4 (14B), Gemma 3 12B, Llama 3.2
Cloud / frontier 70B+ Cloud APIs Claude Opus 4.7, GPT-5.4 Pro, Gemini 3.1 Pro

What 2026 benchmarks actually show

  • Microsoft’s Phi-4 (14B) surpasses many 70B-class models on math, reasoning, and code generation. Training data quality now beats raw parameter count.
  • Qwen 3.5 9B scores 82.5 on MMLU-Pro and 81.7 on GPQA Diamond — comparable to frontier models from 18 months ago.
  • Gemma 3 4B IT hits 71.3% on HumanEval and 89.2% on GSM8K with four billion parameters.
  • Samsung, Google, and Motorola 2026 flagship phones support on-device inference for models up to ~4B parameters.
  • API pricing dropped 40–70% across major providers in 2026, but on-device SLM deployment still delivers 70–90% cost reduction for high-volume predictable workloads.

The hybrid architecture pattern that works in 2026

  1. Route high-volume predictable tasks (classification, extraction, summarization) to local SLMs.
  2. Route complex, novel reasoning to cloud frontier models.
  3. Capture 80–90% of the cost savings while maintaining frontier capability when needed.
Why this matters for your field

If your organization handles sensitive data — health records, student records, financial data, legal documents — you are no longer forced to choose between “use cloud AI and accept compliance risk” or “don’t use AI.” A local 4–9B model running on a single workstation can handle 70-90% of routine tasks without the data leaving your building. This is the single largest unlock for AI adoption in regulated industries.

Trend 4 — Regulation: the EU AI Act is live, and August 2026 is the real deadline

Regulation is the trend most likely to affect your career directly, because the AI literacy obligation is already binding and the main enforcement date is months away.

Current status (April 2026)

Date Status What’s in force
Feb 2, 2025 In force Prohibited AI practices banned. Article 4 AI literacy obligation is active — organizations using AI must ensure staff have sufficient AI literacy.
Aug 2, 2025 In force Governance infrastructure live. Fines active: up to €35M or 7% of global turnover for prohibited practices; €15M / 3% for other violations; €7.5M / 1% for misleading authorities. Obligations for new GPAI models apply.
Aug 2, 2026 Approaching — 3.5 months from this module Main compliance date. Most remaining provisions, including rules for high-risk AI systems (Annex III), become applicable. Enforcement begins at EU and national level. Every Member State must have at least one AI regulatory sandbox.
Aug 2, 2027 Future Rules for high-risk AI embedded in regulated products (toys, medical devices, etc.) apply.

Why Article 4 matters for you specifically

The EU AI Act’s Article 4 — already in force — legally requires organizations deploying AI in the EU to ensure their staff have sufficient AI literacy. Companies operating in the EU are actively hiring people who can run AI literacy training. That’s this course, taught by someone with this course on their resume.

Italy went first

Italy’s Legge 132/2025 (October 2025) became the first EU member state AI framework law with criminal penalties. Fines up to €774,685 plus disqualification from public contracts. A new criminal offense for AI-generated or altered content (e.g., deepfakes) carrying 1–5 years imprisonment. Additional EU member states are expected to follow by the August 2026 main deadline.

Outside the EU

The U.S. approach remains patchwork at the federal level. State-level action is leading:

  • Colorado’s AI Act (effective February 2026) — consumer protection focus
  • California SB 1047 derivatives — frontier model disclosure requirements
  • Texas TRAIGA — workplace AI use disclosure
  • Federal: NIST AI Risk Management Framework (voluntary) — de facto standard for federal contractors

The global direction of travel is clear: AI governance is now a mainstream legal and compliance specialization. The job category “AI compliance officer” barely existed in 2022. It’s a standard Fortune 500 role in 2026.

Trend 5 — Embedded AI: your phone is already an inference server

The least-discussed trend is the most structurally important: AI is becoming invisible. It’s being built into devices and products where you don’t see it — and that’s where most of the actual impact happens.

The 2026 landscape

  • Apple Intelligence (iPhone 15 Pro+, iPhone 16/17, iPad M1+, Mac Apple Silicon) runs on-device inference for a ~3B parameter foundation model with task-specific adapters.
  • Apple-Google Gemini partnership (January 2026): Apple will incorporate Google’s Gemini models into Apple Foundation Models, with on-device distillation for Siri and on-device features. Apple is distilling smaller task-specific models from the full Gemini model with “complete access” to Gemini’s internal reasoning.
  • Gemini Nano runs on Android 16 via the AICore system service on Pixel 8/9 series, Samsung Galaxy S24+, and any Snapdragon 8 Gen 3+ device. It handles summarization, smart replies, rewriting, and translation without cloud round-trips.
  • Cars increasingly ship with on-device AI for natural-language commands, driver monitoring, and predictive maintenance.
  • Smart home devices (Amazon Echo, Google Nest) ship with on-device speech recognition — no more “wake word then cloud.”
  • Healthcare devices, hearing aids, wearables — on-device AI is invading product categories that never had internet connectivity.

Why this is the most important long-run trend

When AI is embedded in a device, three things happen:

  1. Privacy is preserved by default. Data doesn’t leave the device.
  2. Latency drops to zero. No network round-trip means AI feels instant.
  3. AI literacy stops being optional. Every product your organization ships will eventually have AI in it, whether or not your team asked for it.
The implication for your career

If you work in product, design, marketing, legal, compliance, or customer service — and your product or service has any digital component — assume AI is coming to it within 2–4 years whether or not it’s planned. Planning for “AI-adjacent” work in your role is no longer career-specific advice. It’s universal advice.

Augment vs. automate — what’s actually happening

The best data on AI’s real effect on work comes from the Anthropic Economic Index, which analyzes anonymized Claude usage against the U.S. Department of Labor’s O*NET database of 900+ occupations. Five iterations have been published (February 2025 through March 2026).

Key findings across the five reports

Finding Data
Augmentation (AI collaborates with human) ~57% (V1) → ~52–57% (stable)
Automation (AI replaces human on task) ~43% (V1) → ~43–48% (stable)
Occupations with AI in ≥25% of tasks 36% (V1) → 49% (March 2026)
Occupations with AI in ≥75% of tasks 4%
Directively automated tasks (API/business users) 27% (early 2025) → 39% (late 2025)

What the data says, in plain English

  • The augmentation/automation ratio is roughly stable around 55/45. This is not a hockey stick toward automation. Augmentation is the larger category.
  • BUT there is a meaningful divergence: API/business users automate more than consumer users do, and the gap is widening. Enterprise automation grew from 27% to 39% in nine months.
  • Mid-to-high wage occupations use AI most (computer programmers, data scientists, writers, educators).
  • The lowest-paid and highest-paid occupations use AI the least. The middle of the wage distribution is where most action is.

The practical framing

If you’re in a mid-wage knowledge-work role, AI will augment most of your tasks and automate some. The question is not “will AI take my job?” — the data says augmentation dominates. The question is:

“Which of my tasks will be augmented first, and what new tasks will I take on with the time AI saves?”

That is a question you can actually answer for your own job this week. It is also the exact question your future employer will ask in an interview.

Hype vs. substance: a framework you can actually use

Here is the test I apply to every AI claim I see. Four questions. If you can answer all four with specifics, it’s probably substance. If you can’t, it’s probably hype.

The HYPE × SUBSTANCE framework

Question What a substance claim sounds like What a hype claim sounds like
1. Who is paying for this? “Enterprise customer X paid $Y to deploy this in Q3 2025.” “AI is transforming business.”
2. What specifically changed? “Error rate dropped from 8% to 1.5% on Benchmark X.” “AI is getting much better.”
3. Who measured it, independently? “Paper Z, peer-reviewed, replicated by Lab Q.” “According to the company that built it…”
4. What did it get wrong? “Failure mode: hallucinates dates for events after 2023.” “This is a breakthrough.”

The MIT NANDA stress test

Apply the framework to this claim: “Generative AI is transforming enterprise business.”

Who is paying? Enterprises — $30–40 billion invested (MIT NANDA). What changed? 95% of pilots show zero measurable P&L impact. Only 5% achieve rapid revenue acceleration. Who measured it? MIT’s NANDA initiative — 300+ initiative reviews, 52 interviews, 153 survey responses (July 2025). What’s going wrong? The “learning gap” — integration and organizational adaptation, not model quality. Purchased/partnered solutions succeed 67% of the time; internal builds succeed 33% of the time.

Conclusion: The claim is hype in the aggregate, substance at the margins. The 5% that succeed are real. They’re just not 95% of the market. This is the difference between “AI is transforming enterprise” (hype) and “AI transformation is concentrated in 5% of enterprises, characterized by back-office automation deployed through external vendor partnerships” (substance).

This is the level of specificity you want to be able to produce about any AI trend you care about.

The practical interview skill

The employers who will pay the 28% salary premium from Module 12 are not the 95% that are stuck. They’re the 5% who figured out how to deploy AI to move actual numbers. That distinction — the difference between “we’re piloting AI” and “we deployed AI and it saved us $X” — is the single most important signal you can learn to detect in a job interview.

If an interviewer can tell you specifically what their AI deployment did, you’re probably talking to a company in the 5%. If they speak in generalities, you may be looking at 95%.

The 30-minute weekly learning routine

AI literacy decays. The field changes monthly. The only defense is a ritual.

Weekly AI maintenance (30 minutes — same time every week, forever):

  Minutes 0–10:   READ
                  Simon Willison's blog (simonwillison.net)
                  — daily posts from a working practitioner,
                  concrete, almost zero hype

  Minutes 10–20:  PRIMARY SOURCE
                  One post from Anthropic Research,
                  OpenAI Blog, or Google AI Blog —
                  whichever provider your main tool is from.
                  Read the original. Not the summary.

  Minutes 20–30:  HANDS-ON
                  Try ONE new thing in your actual workflow.
                  A new prompting technique, a tool feature,
                  a small experiment. Must be hands-on.
                  Write down one sentence about what you learned.

Do this 50 weeks a year. That's 25 hours of deliberate AI
practice annually. Most AI workers do less.

What to deliberately skip

You will save hours per week by skipping:

  • Twitter/X AI threads. High noise, low signal. The best stuff filters through Simon Willison or Hacker News within 48 hours anyway.
  • LinkedIn thought leadership posts. The ratio of “AI is changing everything” posts to actual information is roughly 100:1.
  • AI newsletter subscriptions beyond two. Pick two. Any more and they’ll pile up and you’ll read none.
  • Model benchmark leaderboard drama. Benchmarks change monthly. If you care about benchmarks, check them once a quarter, not once a week.
  • AGI timeline debates. Unfalsifiable, ages badly, and has zero effect on what you should do Monday morning.

Lab 13.1 — The hype/substance analysis

Pick an AI-related claim you’ve seen in the last week. Could be a headline, LinkedIn post, marketing email, news article — anything.

Run it through the four-question framework:

  1. Who is paying for this?
  2. What specifically changed?
  3. Who measured it, independently?
  4. What did it get wrong?

For each question, write either:

  • A specific fact with a source, OR
  • “I cannot determine this from the claim as presented.”

Deliverable: A one-page analysis with your verdict (Hype / Mostly Hype / Mostly Substance / Substance) and one sentence explaining what would change your verdict.

The goal is not to be cynical about AI. The goal is to build the reflex to ask the four questions automatically before you form an opinion on any new AI claim.

Lab 13.2 — Your first on-device AI

If you have an iPhone 15 Pro or newer, or a Pixel 8 or newer, or a Galaxy S24 or newer, your phone already has a multi-billion-parameter language model running on-device.

Task: Find one built-in AI feature on your phone that runs on-device (not cloud). Use it for one real task this week. Write a 200-word reflection covering:

  1. What feature you used.
  2. What it did well.
  3. What it did poorly.
  4. How you would know, without reading documentation, whether it runs on-device or in the cloud.

(Hint for item 4: airplane mode. If the feature works in airplane mode, it’s on-device.)

Exercise 13.3 — Future-of-your-job analysis

Pick a job title you might realistically hold in 5 years. (Not a fantasy role. A realistic next step.)

Using the Anthropic Economic Index data and the Lightcast career-area data from Module 12:

  1. Adoption stage: What percentage of postings for this role mention AI skills today? (Use Module 12’s table.)
  2. Augmentation estimate: Based on the 57/43 augmentation/automation split and what you know about the role, which of its tasks are most likely to be augmented first?
  3. Automation risk: Which tasks, if any, look like candidates for full automation?
  4. New tasks: With the time saved, what higher-value tasks would the role shift toward?
  5. Your move: Given the above, what specific skill should you develop in the next 12 months?

Deliverable: A one-page memo written to your future self.

Quiz

1. As of April 2026, the Anthropic Economic Index finds that the augmentation/automation ratio across Claude usage is approximately:

  1. 90% augmentation / 10% automation
  2. 57% augmentation / 43% automation, roughly stable
  3. 25% augmentation / 75% automation
  4. Shifting rapidly toward automation

2. The MIT NANDA 2025 report found that what percentage of enterprise generative AI pilots deliver measurable P&L impact?

  1. 5%
  2. 25%
  3. 50%
  4. 95%

3. The EU AI Act’s Article 4 (AI literacy obligation) is currently:

  1. Not yet in force
  2. In force since February 2025 and binding on organizations deploying AI in the EU
  3. In force only for high-risk AI systems
  4. Applicable only to EU-headquartered companies

4. The main enforcement date for most remaining EU AI Act provisions is:

  1. February 2025
  2. August 2025
  3. August 2026
  4. August 2030

5. Which of the following is NOT a valid application of the hype/substance framework?

  1. Asking who is paying for the AI deployment
  2. Asking who measured the outcome independently
  3. Asking whether the CEO is excited about AI
  4. Asking what the AI got wrong

6. The current leading standalone text-to-video models in April 2026 are:

  1. Sora and Sora Pro
  2. Veo 3.1, Kling 3.0, and Seedance 2.0
  3. DALL-E 4 and Midjourney
  4. Stable Diffusion XL

7. The best evidence-based recommendation for staying current with AI is:

  1. Subscribe to 10+ AI newsletters
  2. Follow AI influencers on LinkedIn and Twitter
  3. A 30-minute weekly routine with one primary-source reading and one hands-on experiment
  4. Take a new certification every 6 months

8. If you work in a mid-wage knowledge-work role, the most useful question to ask about your own future is:

  1. “Will AI replace my job?”
  2. “When will AGI arrive?”
  3. “Which of my tasks will be augmented first, and what new tasks will I take on with the time AI saves?”
  4. “Should I switch careers now?”

(Answers: 1-b, 2-a, 3-b, 4-c, 5-c, 6-b, 7-c, 8-c)

✅ Test Yourself: AI Futures
  1. Name the five trends covered in this module. (Agents, multimodal, small models, regulation, embedded AI)
  2. What does the “jagged technological frontier” mean for your work? (AI is uneven — some tasks it handles brilliantly, adjacent tasks it fails at. You need to know the boundary.)
  3. Why is a 30-minute weekly routine better than 5 hours of AI news scrolling? (Signal over noise. Curated sources with specific learning goals beat unfocused consumption.)
Guided Practice 13.1 — The Hype Detector

Pick any AI headline from the last 7 days. Apply the 5-filter test: (1) Peer-reviewed evidence? (2) Meaningful sample size? (3) Specific model/version? (4) Limitations acknowledged? (5) Does the claim survive restating without “AI”? Write a 3-sentence verdict: hype, substance, or too early to tell.

Module 13 Summary

Key Takeaways
  1. Every AI prediction is wrong — build judgment, not certainty
  2. Five trends matter now: agents, multimodal, small models, regulation, embedded AI
  3. Augmentation dominates automation — AI changes roles, rarely eliminates them
  4. The hype vs. substance framework is more valuable than any specific prediction
  5. 30 minutes per week with curated sources beats hours of unfocused AI news
💭 THINK ABOUT IT

Every generation thinks its technology changes everything. The printing press. The telegraph. Electricity. The internet. They were all right — eventually. But the people who benefited most weren’t the ones who predicted the future correctly. They were the ones who built useful things with the technology while everyone else was still debating whether it would last. You now know enough to build. The question is whether you will.

🔬 Lab 13: Design Your 30-Day AI Challenge

Lab 13: Design Your 30-Day AI Challenge

Time: 45 minutes | Tools needed: Any AI chatbot + a planning template

Purpose: Create a structured 30-day plan to deepen one specific AI skill after this course ends.

Instructions: 1. Pick one skill from this course to develop further 2. Design 30 daily tasks (5-15 minutes each) that build progressively 3. Include 4 checkpoints (days 7, 14, 21, 30) with self-evaluation criteria 4. Define success — what does mastery look like after 30 days?

Coming up in Module 14

You’ve built the knowledge. Now you demonstrate it. Module 14 is your capstone.

Previous: Module 12 — AI and Your Career | Next: Module 14 — Capstone Project