top of page
Search

How Chain-of-Draft reasoning taught machines to keep it short and sweet

  • Writer: Vishisht Choudhary
    Vishisht Choudhary
  • Aug 7
  • 5 min read

Updated: Aug 12


The Problem: When Your AI Won't Stop Talking


Picture this: You ask an AI a simple math question like "What's 20 minus 12?" and instead of just saying "8," it launches into a full dissertation, completely overshooting on the intended task.

Sound familiar? Welcome to the wonderful world of Chain-of-Thought (CoT) reasoning, where AI models just don’t know when to shut up. I seemed to be getting this problem especially when using models such as Deepsek’s R1. I would ask a simple question and it would reason on and on contradicting itself, just to give me a wrong answer


Don't get me wrong: CoT can be incredibly useful for complex reasoning tasks. But sometimes you just want the AI equivalent of a quick "yep" instead of a TED talk about why the sky is blue.


ree

Enter Chain of Draft: The AI That Actually Learned to Edit

Chain of Draft (CoD) reasoning is like giving your AI a really good editor who's not afraid to use the delete key. Instead of showing you every single thought bubble (looking at you, CoT), CoD works more like how humans actually solve problems: sketch it out, refine it, polish it up, and then show the final result.

Here is an example to illustrate how big the difference is:


Scenario:

"A company is considering launching a new product. The development cost is $500,000, expected revenue in year 1 is $300,000, and year 2 is $450,000. Should they proceed?"


Standard Prompt

Input:

A company is considering launching a new product. The development cost is $500,000, expected revenue in year 1 is $300,000, and year 2 is $450,000. Should they proceed with the launch?

Output:

Based on the financial projections, the company should proceed with caution. The total revenue over two years ($750,000) exceeds the development cost ($500,000) by $250,000, indicating a positive return. However, the company won't break even until sometime in year 2, which represents significant risk. The decision should also consider market conditions, competition, and the company's cash flow capacity to sustain losses in year 1.

Chain of Thought Prompt

Input:

A company is considering launching a new product. The development cost is $500,000, expected revenue in year 1 is $300,000, and year 2 is $450,000. Should they proceed with the launch?

Please analyze this step by step, considering:
1. The financial implications
2. Break-even analysis
3. Risk factors
4. Your final recommendation with reasoning

Output:

Let me work through this business decision systematically:

1. Financial Implications:
First, let me calculate the total investment and returns. The upfront development cost is $500,000. In year 1, they expect $300,000 in revenue, which means they'll still be $200,000 short of breaking even ($500,000 - $300,000 = $200,000 remaining). In year 2, they expect an additional $450,000 in revenue. So the total revenue over two years would be $300,000 + $450,000 = $750,000.
															...Bla Bla Bla...

4. Final Recommendation:
Based on this analysis, I would recommend proceeding with the launch, but with important caveats. The financial projections show a reasonable return on investment with $250,000 profit over two years (50% ROI). However, the company must ensure they have adequate cash reserves to cover the initial $200,000 shortfall in year 1 and any unforeseen costs. They should also develop contingency plans in case revenue projections fall short and consider ways to accelerate revenue or reduce costs. Additionally, they should validate their revenue assumptions through market research and pilot testing before full commitment.

Chain of Draft Prompt


Input:

A company is considering launching a new product. The development cost is $500,000, expected revenue in year 1 is $300,000, and year 2 is $450,000. Should they proceed with the launch?

Draft your analysis in minimal steps (max 5 words each). Focus on: costs, revenues, break-even, decision. Final recommendation after ####.

Output:

Cost analysis: $500k upfront investment
Revenue projection: $750k total return
Break-even timeline: Mid-year two approximately
Net outcome: $250k profit potential
Risk assessment: Cash flow concerns

#### PROCEED WITH CAUTION. Positive ROI ($250k profit over 2 years) but requires strong cash reserves for 18-month break-even period. Validate revenue assumptions before full commitment.

Token Usage Comparison

Approach

Word Count

Estimated Tokens

Efficiency

Standard

68 words

~85 tokens

Baseline

Chain of Thought

412 words

~515 tokens

6x more tokens

Chain of Draft

39 words

~50 tokens

41% fewer tokens


The Magic Numbers (That'll Make Your CFO Very Happy)

Here's where things get exciting for anyone who's ever had to explain API costs to their boss:

  • Token reduction: CoD uses as little as 7.6% of the tokens compared to CoT (that's like going from a novel to a haiku)

  • Speed boost: Up to 76% latency reduction (faster than your colleague running to the break room when free donuts are announced)

  • Cost savings: 40% reduction in total tokens used (your budget will thank you)

Essentially, CoD is the Marie Kondo of AI reasoning—it keeps only what sparks computational joy and tosses the rest.


The Beautiful Marriage: CoD + Low-Code = Magic

And of course it is possible to marry the idea of Chain of Draft with Low Code Automations.

Here's how this technological love story unfolds:

  1. Trigger: Something happens (new email, form submission)

  2. Draft Generation: Your AI creates a quick, efficient first draft using CoD

  3. Refinement Loop: The low-code platform orchestrates multiple rounds of AI refinement (because even AI needs feedback)

  4. Final Action: The polished result gets sent, saved, or deployed

Real-World Superpowers You Can Actually Use

Content Creation: Transform rambling voice notes into LinkedIn posts.

Customer Service: Deploy AI chatbots that actually help customers instead of making them want to throw their phones across the room.

Data Analysis: Turn spreadsheets full of incomprehensible data into insights that even your least technical colleague can understand.

Approval Processes: Automate the bureaucratic maze of getting anything approved in your organization (results may vary depending on organizational stubbornness levels).


Best Practices


Choose Your Weapons Wisely

  • For Beginners: Start with Zapier, it's like training wheels for automation

  • For Business Users: Make.com offers more flexibility without requiring a computer science degree

  • For Technical Teams: n8n gives you all the power and control you could want


Craft Your Prompts Like a Pro

Remember, with CoD, less is more. Your prompts should be like a good tweet: concise, clear, and to the point. Include examples, specify your desired output format, and don't be afraid to tell the AI to keep it short and sweet.

Monitor and Improve

Set up monitoring like you're tracking your favorite sports team's stats. Watch for performance metrics, error rates, and user feedback. The beauty of this approach is that you can continuously refine and improve your workflows.


The Bottom Line


Chain of Draft reasoning combined with low-code automation is like giving every business professional a superpower. It's the difference between spending your day on repetitive tasks versus spending it on work that actually moves the needle. The future belongs to organizations that can adapt quickly, think creatively, and leverage technology to do more with less. And thanks to CoD and low-code platforms, that future is available today.



About NOA

NOA is your expert for process automation and custom tooling. Our solutions begin where standard software ends. We combine market-leading low-code tools to create robust end-to-end automations that grow with your requirements and actively support your business growth.

Our approach eliminates the classic disadvantages of low-code solutions while leveraging their major benefits across the board. The result: the reliability and scalability of a perfect standard tool, developed exclusively for your needs.

Most importantly, we transform your teams – including non-technical employees – into true automation heroes. This way, we create a sustainable "Automation First" culture for your long-term competitive advantage.

bottom of page