Forem: Richard Ketelsen

Introducing CRAFT for Cowork

Richard Ketelsen — Tue, 31 Mar 2026 12:36:45 +0000

Claude Cowork is stateless. Every session starts blank — no project memory, no context from yesterday, no record of the decisions you already made.

If you've built anything complex with it, you know the workaround: paste a wall of context at the start of every session and hope nothing gets lost. That approach breaks down fast.

CRAFT (Configurable Reusable AI Framework Technology) is a structured communication framework that fixes this. It provides 188 recipes across 12 cookbooks that handle session continuity, automatic git backup, quality verification, and multi-persona collaboration — all running inside Cowork's native environment.

What the recipes cover

Session handoffs: Structured snapshots capture your full project state and restore it next session. Tested across 72 development sessions — 71 consecutive handoffs with zero data loss, surviving context compaction events, crashes, and device switches.

Git integration: Event-driven commits (not timer-based) that protect every change. Cowork's Linux sandbox resets between sessions; CRAFT ensures your work is committed before that happens.

Quality assurance: A 14-point QA plan with 4 verification gates — file pointability, read-vs-reconstructed detection, constraint conflict checking, and untested assumption flagging. During development, this caught a licensing error across 9 content files before publication.

Multi-persona system: Separate AI personas for different roles (operations, content creation, validation). The Creator/Validator workflow prevents the AI from reviewing its own output.

The numbers

188 recipes, 12 cookbooks (from ~60 pre-project — 213% growth)
7 formal specifications
64 documented lessons learned
~59,000 lines of recipe content
9 projects in a unified monorepo

Get it

CRAFT for Cowork is available now as a free public beta:

GitHub: github.com/CRAFTFramework/craft-framework

The Language Spec is licensed under BSL 1.1 (converts to Apache 2.0, January 1, 2029). Content is proprietary via membership. Full license details: craftframework.ai/craft-license

This is week 1 of an 8-week series. Next week: how the handoff system eliminates the session amnesia problem.

Session Amnesia: The Hidden Cost of Multi-Session AI Work

Richard Ketelsen — Tue, 24 Mar 2026 00:44:49 +0000

If you use Claude's desktop app — Cowork — for anything beyond single-shot tasks, you've hit this wall.

Every time you close a session and open a new one, the AI starts fresh. No memory of what you discussed. No knowledge of decisions you made. No context from yesterday's work. You re-explain everything from scratch.

The math nobody talks about

Even 15 minutes of context re-entry per session adds up. Three sessions a day? That's nearly 4 hours a week just re-explaining things the AI already knew yesterday.

And it's not just the reset between sessions. Within a single long session, Cowork automatically summarizes earlier parts of the conversation as it approaches the context limit. Decisions you made in the first hour may be compressed to a single sentence by hour three.

The workarounds are expensive too

Most people adapt with one of three patterns:

Copy-paste context. Paste your notes at the start of each session. Works, but takes 20-40 minutes per session. On a team, that's hours per week.

Mega-sessions. Keep everything in one long conversation. Avoids resets but degrades over time — earlier context gets summarized, outputs drift from earlier decisions.

Living documents. Maintain a Google Doc or Notion page as the AI's "memory." Better, but now you're maintaining another system alongside the AI itself.

None of these are elegant. They're all working around a limitation instead of working with the tool.

Why it matters

The 77% of workers in Upwork's 2024 study who said AI increased their workload? This is a contributing factor. The friction didn't disappear — it moved from "doing the work" to "preparing to do the work."

The AI itself is fast and capable. The session architecture isn't built for continuous work. That gap is where the efficiency loss lives.

The real question

Session amnesia isn't a bug — it's a design constraint of the desktop environment. But design constraints can be addressed.

If you're doing multi-session work in Cowork or similar tools: how are you handling context continuity? What's working, what's not?

Curious to hear what patterns people have found.

Session Amnesia: The Hidden Cost of Multi-Session AI Work

Richard Ketelsen — Mon, 16 Mar 2026 19:11:04 +0000

If you use Claude's desktop app — Cowork — for anything beyond single-shot tasks, you've hit this wall.

The math nobody talks about

Even 15 minutes of context re-entry per session adds up. Three sessions a day? That's 5+ hours a week just re-explaining things the AI already knew yesterday.

The workarounds are expensive too

Most people adapt with one of three patterns:

Copy-paste context. Paste your notes at the start of each session. Works, but takes 20-40 minutes per session. On a team, that's hours per week.

Mega-sessions. Keep everything in one long conversation. Avoids resets but degrades over time — earlier context gets summarized, outputs drift from earlier decisions.

Living documents. Maintain a Google Doc or Notion page as the AI's "memory." Better, but now you're maintaining another system alongside the AI itself.

None of these are elegant. They're all working around a limitation instead of working with the tool.

Why it matters

The AI itself is fast and capable. The session architecture isn't built for continuous work. That gap is where the efficiency loss lives.

The real question

Session amnesia isn't a bug — it's a design constraint of the desktop environment. But design constraints can be addressed.

If you're doing multi-session work in Cowork or similar tools: how are you handling context continuity? What's working, what's not?

Curious to hear what patterns people have found.

78% of AI Users Bring Unapproved Tools to Work — There's a Structural Fix

Richard Ketelsen — Mon, 23 Feb 2026 21:45:02 +0000

The Data

Two independent studies found the same number in 2024-2025:

Microsoft/LinkedIn 2024 Work Trend Index: 31,000 knowledge workers, 31 countries, conducted by Edelman Data & Intelligence. Finding: 75% use AI at work. Of those, 78% use unapproved tools (BYOAI).
WalkMe 2025 AI in the Workplace Survey: 1,000 U.S. working adults who use AI, conducted by Propeller Insights. Finding: 78% admit to using AI tools their employer never approved.

Same percentage. Independent methodologies. Different sample populations. The convergence is what makes this credible — two unrelated research efforts produced identical results.

Why It Matters for Developers

Every unapproved AI interaction is a data handling question. When a developer pastes proprietary code into an unvetted AI chatbot, that code hits someone else's infrastructure under someone else's data retention policy. When a PM uploads a roadmap to a free AI tool, that document exists on servers your organization never evaluated.

For anyone working in regulated industries or with sensitive intellectual property, the risk compounds. Most free AI tools have terms of service that permit using inputs for model training. Some retain data indefinitely. The specific policies vary, but the common problem is clear: you don't control what happens to the data after it leaves your machine.

IBM's 2024 Cost of a Data Breach Report found shadow AI adds an average of $670,000 to breach costs. That's not the breach itself — that's the additional cost from the unapproved AI component alone.

The Structural Problem

The pattern repeats across organizations:

1. Organization approves AI Tool X
2. Tool X handles ~60% of workflows
3. Employees find Tools Y, Z for the rest
4. Tools Y, Z are SaaS platforms with own data policies
5. IT has no visibility into Tools Y, Z
6. Data leaks through channels that don't exist in the threat model

Enforcement doesn't scale when 78% of AI users are finding alternatives. You can write policies faster than employees can find workarounds, but the underlying incentive — getting work done efficiently — always wins.

The problem isn't people. It's architecture. Centralized AI platforms create a gap between approved capability and actual need. Employees fill that gap with whatever's available.

A File-Based Alternative

CRAFT Framework approaches this differently. Instead of another platform to approve, evaluate, and monitor, CRAFT stores AI workflows as plain text files on your own machine:

~/craft-projects/
├── recipes/          # Reusable prompt templates
├── cookbooks/        # Organized recipe collections
├── projects/
│   ├── project-001/
│   │   ├── chat-history.txt
│   │   ├── project-config.txt
│   │   └── handoffs/
│   └── project-002/
└── backups/          # cp -r. That's it.

No platform. No installation. No vendor to vet. No approval process. Files work with any AI chat tool — Claude, ChatGPT, Gemini, or the next one. You're not adopting software. You're organizing text files.

The key insight: CRAFT separates the workflow from the tool. Your prompts, templates, and project context are just files. The AI tool is the execution layer. Since the files never leave your machine, there's no shadow IT category to manage and no data going anywhere you don't control.

Backup strategy? cp -r craft-projects/ /backup/drive/. That's the entire disaster recovery plan.

The 78% BYOAI problem exists because centralized platforms create a gap between what's approved and what people need. CRAFT removes the platform from the equation entirely.

Beta is open: craftframework.ai

Sources: Microsoft/LinkedIn 2024 Work Trend Index; WalkMe 2025 AI in the Workplace Survey; IBM 2024 Cost of a Data Breach Report

87% of Orgs Lost SaaS Data Last Year. Is Your AI Workflow Next?

Richard Ketelsen — Mon, 16 Feb 2026 20:01:45 +0000

Two industry surveys. Two different companies. Same finding.

87% of IT professionals reported their organization experienced SaaS data loss in 2024 (Kaseya 2025 Report, 3,000+ respondents).

79% of IT decision-makers incorrectly believed their SaaS apps included backup and recovery by default (Rewind 2024 Report, 419 respondents).

These aren't from the same survey. Two separate companies, surveying different populations at different times, found convergent evidence of the same systemic problem.

The Shared Responsibility Model

If you've worked with cloud infrastructure, you know this one. Your provider guarantees uptime and platform availability. Your data — accidental deletions, overwrites, rogue integrations, insider sabotage — is your problem.

GitHub's Terms of Service explicitly disclaim liability for data loss. So does Microsoft. So does Google. The 79% who assumed otherwise simply didn't read the fine print.

This isn't a niche legal technicality. It's the operational reality of every SaaS platform. The provider replicates data across their infrastructure to protect against hardware failure on their end. If your data gets deleted because of something that happened on your end — human error, a misconfigured integration, a bad script, a malicious actor — that's on you.

The Recovery Gap

Even when organizations detect data loss quickly, recovery is rarely instant:

14% can recover SaaS data within minutes
~40% recover within hours
35% need days or weeks
2% cannot recover at all
Only 40% of IT professionals are confident their backup systems would hold up in an actual crisis

Proofpoint's independent 2024 study of 600 security professionals across 12 countries found 84.7% experienced at least one data loss incident — corroborating the Kaseya findings from an entirely separate source.

What This Means for AI Workflows

Think about where your AI context lives. Your system prompts, your refined instructions, your conversation histories — they're sitting in a SaaS platform you don't control.

Most AI services offer limited or no export features. The ones that do typically dump raw conversation logs — not organized, usable records. One account lockout, one policy change, one provider decision, and you're rebuilding from memory.

If you've ever lost a carefully constructed GPT configuration or had a conversation thread disappear, you know this isn't theoretical.

A Different Architecture

CRAFT Framework stores everything as plain text files on your machine.

~/craft-projects/
├── recipes/          # Reusable AI workflow templates
├── cookbooks/        # Organized recipe collections
├── projects/
│   ├── project-001/
│   │   ├── chat-history.txt
│   │   ├── lessons-learned.txt
│   │   └── handoffs/
│   └── project-002/
└── backups/          # cp -r craft-projects/ /backup/drive/

That's the entire backup strategy. cp -r. No third-party backup service. No API dependency. No subscription to maintain access to your own work.

Because CRAFT files are text — not software, not a database, not a SaaS wrapper — they carry none of the vulnerability profile of a hosted platform:

Can't carry malware — they're plain text
Can't be corrupted by a platform breach — they're on your machine
Can't be held hostage by policy changes — you own the files
Can't be lost if a provider goes down — they never depended on the provider

Your data goes directly from you to whatever AI service you choose. No middleman. No additional attack surface. No Terms of Service disclaiming responsibility for your work.

CRAFT session histories aren't raw dumps either. Each conversation produces a curated summary — decisions, insights, next steps — that's actually useful as a reference document. Back those up and you have a complete, organized record of every AI workflow you've built.

The Market Is Catching Up

Gartner predicts 75% of enterprises will prioritize SaaS backup by 2028, up from just 15% in 2024. The industry is recognizing a problem that CRAFT's architecture solved from day one — not by adding another protection layer on top of fragile infrastructure, but by removing the fragile infrastructure entirely.

Beta is open: craftframework.ai

Sources: Kaseya/Spanning 2025 SaaS Backup Report; Rewind 2024 SaaS Data Report; Proofpoint 2024 Data Loss Landscape Report; Gartner August 2024

Three Pre-Beta Messages That Validated CRAFT Framework

Richard Ketelsen — Tue, 10 Feb 2026 17:13:24 +0000

What Clicked

We ran an 8-week Pre-Beta campaign for CRAFT Framework across every major platform. Three messages clearly outperformed everything else—and together, they tell us something important about what developers and AI power users actually need.

#1: No Code, No Problem

The most popular message wasn't technical. It was anti-technical.

Most people hear "AI framework" and think git clone, npm install, config files. CRAFT challenged that assumption. A Recipe in CRAFT isn't source code—it's a structured conversation pattern stored in a Python-compatible text file. If you can write clear instructions, you can build one.

# This IS a CRAFT Recipe (simplified)
RECIPE = {
    "title": "Code Review Assistant",
    "parameters": ["language", "code_block"],
    "steps": [
        "Analyze the code for bugs and style issues",
        "Suggest improvements with explanations",
        "Rate overall quality on a 1-10 scale"
    ]
}

The barrier to structured AI workflows isn't technical skill. It's the assumption that structure requires technical skill.

#2: The Classification Tax

The privacy message broke through because it reframed the problem.

Stat	Source
57% say privacy is AI's #1 barrier	IBM Global AI Adoption Index 2024
48% share non-public data with AI anyway	Cisco Data Privacy Benchmark 2024
35% of AI-sent data is classified sensitive (3x increase)	Cyberhaven 2025
78% bring their own AI tools without approval	Microsoft Work Trend Index 2024
65% bypass security policies for productivity	CyberArk 2024

We called this the Classification Tax: the cognitive overhead of deciding what's safe to share with AI on every single interaction. CRAFT handles this architecturally—Recipes define data boundaries once, enforced on every execution.

#3: Templates for AI Conversations

What if your best AI conversation could be a template for every future conversation?

This question resonated because every developer has experienced the inverse: the perfect prompt that vanished into chat history. CRAFT Recipes capture the full interaction pattern—prompt, context, persona, variables, execution flow—so your best work becomes a reusable tool.

The Pattern

These three messages validated that CRAFT is solving real problems:

Accessibility — AI productivity tools shouldn't require programming
Privacy by architecture — Classification decisions made once, not per-chat
Reusability — Best AI work preserved and repeatable

Try It

CRAFT Beta launched February 1, 2026. Built on 50+ completed projects. We call early members Founding Chefs—they're writing the first cookbook.

→ craftframework.ai

OOP + AI = CRAFT

Building a Framework Using Itself: The CRAFT Meta-Proof

Richard Ketelsen — Mon, 02 Feb 2026 14:25:38 +0000

The Challenge

Eight weeks ago, I had work from the CRAFT Alpha and a clear vision. No website, no brand, no promotional infrastructure.

I also had a constraint: I work full-time Wednesday through Saturday in cybersecurity. CRAFT only gets Sundays and Mondays.

The Approach

CRAFT (Configurable Reusable AI Framework Technology) applies OOP principles to AI conversations. The core insight: treat AI sessions like objects with state persistence, not isolated function calls.

So I used CRAFT to build CRAFT.

The Implementation

Six projects handled everything:

CFT-PROJ-CP-045 (Branding): Created complete brand identity in 2 days using the Branding and Identity Cookbook.

CFT-PROJ-CP-044 (Tech Support): Attached config files, now it answers "how do I do X?" with steps tailored to my actual setup.

CFT-PROJ-CP-047 (Version Control): Teaches me Git in small words. Knows my skill level.

CFT-PROJ-CP-046 (Automation): Guides automation based on my specific environment.

CFT-PROJ-CP-048 (Promotion): Created this entire marketing campaign—72 pieces of content across 9 platforms, including this article. The marketing promoting CRAFT was built using CRAFT.

CFT-PROJ-CP-049 (Master): Connects 56 past chats—the most important information from ~28,000 pages of history. The Chat History file is 34,915 lines.

The Key Insight

The magic isn't in the AI model. Everyone has access to Claude, GPT-4, Gemini.

The magic is in structure. At each session's end, CRAFT captures only the most important information—key decisions, critical learnings, what worked, what didn't.

The next session doesn't get a data dump. It gets distilled knowledge.

Imagine an AI that actually remembers everything important about your work.

The Result

14 working days. Complete brand, functional website, 8-week marketing campaign, and a framework sophisticated enough to manage itself.

I'm not a professional developer. Haven't been a designer in decades. If I can build this with weekend availability, what could you build with your expertise?

Try It

The CRAFT Beta launched yesterday. Founding Chefs get free access and permanent early adopter status.

https://craftframework.ai/register/founding-chef/

No Code? No Problem. CRAFT Beta Opens in 6 Days

Richard Ketelsen — Mon, 26 Jan 2026 20:23:14 +0000

CRAFT applies OOP principles to AI workflows. But here's what might surprise you: you don't need to code to use it.

The Programming Concepts (Without the Programming)

Recipes = Reusable AI conversation templates
Cookbooks = Organized collections of recipes
Handoffs = Context that carries between sessions
Variables = Placeholders for information that changes

If you can follow a cooking recipe and organize files into folders, you already understand these concepts.

Who CRAFT Is For

Content creators wanting consistent AI outputs
Business professionals using AI for reports and analysis
Consultants needing repeatable workflows
Developers (of course)
Anyone who uses AI regularly and wants better results

What Beta Members Get

Ready-to-use recipe library
Simple guides for creating custom recipes
Founding Chef permanent recognition
Welcoming community
Privacy by design (self-hosted WordPress)

The Technical Foundation

For the devs who want to know: CRAFT runs on self-hosted WordPress. Your data stays on your infrastructure. No third-party processing of your prompts.

But you don't need to understand the architecture to use the recipes.

Beta opens February 1, 2026.

Developers welcome. Non-developers equally welcome.

No code? No problem.

CRAFTFramework.ai 🍳

13 Days: CRAFT Beta Final Countdown

Richard Ketelsen — Mon, 19 Jan 2026 23:20:59 +0000

13 Days: CRAFT Beta Final Countdown

8 weeks of Pre-Beta content. 70+ weeks of solo development. February 1, 2026: Beta opens.

The Technical Foundation

CRAFT applies OOP principles to AI workflows:

Recipes = Reusable procedures (functions)
Cookbooks = Organized collections (classes)
Handoffs = Session-to-session state management
PROJECT_VARIABLES = Persistent configuration

Self-hosted WordPress architecture = privacy by design.

Beta Member Benefits

Founding Chef permanent recognition
48 projects of battle-tested recipes
Direct influence on development
Path to Beta Legend (free lifetime for exceptional contributors)

The Honest Reality

This is a Beta. Documentation has gaps. Community starts from zero.

If you want polished, wait. If you want to help build—13 days.

CRAFTFramework.ai

Final week of Pre-Beta campaign. Thanks for following along.

20 Days Until CRAFT Beta: OOP Principles for AI Workflows

Richard Ketelsen — Mon, 12 Jan 2026 23:35:03 +0000

CRAFT Framework applies object-oriented programming concepts to AI conversations:

Recipes = Reusable procedures (like functions)
Cookbooks = Organized collections (like classes)
Handoffs = Session continuity (like state management)
PROJECT_VARIABLES = Persistent configuration

After 70+ weeks of solo development across 48 projects, Beta opens February 1, 2026.

Beta members get:

Founding Chef permanent recognition
Direct influence on development
Path to Beta Legend status (free lifetime for exceptional contributors)
Privacy by architecture (self-hosted WordPress)

Full details: CRAFTFramework.ai

Building in public: this is the penultimate week of Pre-Beta content before launch.

Behind the Scenes: Building CRAFT Solo

Richard Ketelsen — Mon, 05 Jan 2026 21:40:01 +0000

The Constraint

Full-time job Wed-Sat. CRAFT development Sun-Mon. 48 projects solo.

Why It Matters

If CRAFT helps accomplish a week's work in one day, the methodology works. The constraint is the test.

The Unknowns

→ How patterns translate to different industries
→ Right complexity for new users
→ What I've adapted to but can't see
Beta opens February 1 at craftframework.ai

Build Your First AI Recipe in 5 Minutes

Richard Ketelsen — Mon, 29 Dec 2025 22:23:05 +0000

The Misconception

CRAFT uses programming metaphors. Variables. Functions. Objects.
But you don't need to code. These are structural concepts, not implementation details.

Start Here

Pick a task you repeat with AI
Answer: What to do? What context? What output?
Identify what changes vs. what stays constant That's your first recipe.

Learn more at craftframework.ai