🦾Eric Nowoslawski

Growth Engine X

Founder

Clay

Certified Expert

Smartlead

Instantly.ai

REITIUM

Advisor

Gated

Advisor to an exciting Startup called Gated. Gated challenges unknown senders to donate to charity in order to reach you - so everything in your inbox is worth your attention.

It's like a postage stamp for email. 

As a huge email sender, I'm most excited for the ability of our messages to cut through the noise, prove to our prospects that we are being intentional and we want to solve their problems, instead of relying on fancy copywriting and sending 10s of thousands of emails.

Gateway Regional Chamber of Commerce

Vice President

Help Chamber Members grow their business
Facilitated Connecting Members through observed Synergies
Public Policy Research and Advocacy
Negotiate discounts on behalf of the Chamber
Engaged with Local Municipalities in Economic Development and Business Policy

The Gateway Regional Chamber of Commerce is the largest business association in the New York Metropolitan Area with about 1,200 members. We hold approximately 150 events per year from large 600-person dinners to small 10-person breakfasts, and everything in between, with a great emphasis on networking and business promotion.

The Gateway Chamber has a long history dating back to 1911 when it was founded as the Elizabeth Chamber of Commerce by John D. Rockefeller and other New Jersey industrialists.

Today, the Gateway Chamber serves as an umbrella organization for a number of regional and local chambers and networking groups including: the Union County Chamber of Commerce; the Central Jersey Chamber of Commerce; the Route 22 Chamber of Commerce; the Kenilworth Chamber of Commerce; the Roselle Park Chamber of Commerce; the Clark Chamber of Commerce; the Linden Chamber of Commerce, the Somerset Hills Business Network; the Black Business Alliance; the Gateway Warehouse Association; the Gateway Hospitality Alliance; and The Irish Business Association. When you join the Gateway Chamber you automatically become a member of ALL these groups at no additional cost.

The Gateway Chamber is a strong advocate for its members. It works at the federal, state and local levels on both legislative and regulatory issues.

We consider it to be core to our mission to help our members grow their businesses, and as a result our retention rate has averaged 95% over the last five years. 

Members of the Gateway Chamber run the gamut from large corporations like Merck, Comcast, Verizon and Continental Airlines, to small and medium-sized business, to a very active home-based business group.

Director of Special Projects

CCFGiving

Operations Associate

Eastern National Academy

Instructed, refined, and improved skills for gymnasts of all levels
• Earned a 100% retention rate for all Recreational classes 3 months after starting
• Created Ninja Warrior Program that retained 180% of students over 9 months
• Manage employees and planning for Birthday parties and special events

Men's Gymnastics Coach

I am a direct response outbound lead generation expert. And by that jumble of words, I mean that I leverage cold email to build predictable revenue pipelines for my clients. 

I leverage over 20 databases to merge Firmographic, Technographic, and Persona data to build highly targeted playbooks with my clients. 

I believe we need to arm sales teams with the knowledge of their total Target Addressable Market and teach them how to score that Market for which companies have the most pain and need their solution. 

The best way to do this is to look at public indicators such as open jobs data, employee growth/attrition, social media growth, web traffic, new product launches, employee sentiment, individual employee tech investment, company tech stacks, etc.

With this data, not only are you only messaging with people that you know have a problem that you can solve, you also are armed  with powerful messaging that cuts through the noise. 

I teach sales teams to embrace technology and multiply what is possible per employee.

Instantly.ai just released something I've even gone as far to build for myself. But it's way more powerful because of their scale.

They just released their upgraded email verification enrichment for catch all/risky email addresses. 

This is huge for a company like them to be doing because they are using their scale to be able to verify the email addresses with real sending instead of an SMTP check or other methods of catch all verification. 

Now users will be able to check if anyone else on the platform has sent an email to the email address you uploaded and see if it went through successfully in the past. 

Allowing you to contact people that usually don't get as many emails because platforms have a tough time with catch all verification. 

We tried to build this based on our own sending data too but it just didn't have the scale necessary to be useful. With the amount of emails Instantly sends on a daily basis, it's incredibly valuable data for those serious about keeping their bounce rates down. 

I don't think this was publicly announced either but I've noticed that if you upload a lead list, they are automatically taking out leads that have a history of bouncing 100% of the time and I haven't even been charged for that. 

Very interesting development for them to keep pushing features for the cold email community.

Founder Growth Engine X | Clay Enterprise Partner

I've been completely wrong about AI prompting, and it's cost me thousands of dollars and it might be costing you money too. 

For months, I've been telling my team (and all of you) to structure prompts like this:

1. Tell the AI what you want it to do
2. Give it the data to analyze and reference
3. Provide examples of good outputs

I didn't realize that this approach wastes tokens. Especially when you're running outbound campaigns at scale.

I know this may seem obvious to technical people out there but I didn't even consider making a change. 

I had a call with my enterprise rep from OpenAI last week and they asked me why I wasn't using cached inputs based on my use case. 

I didn't even consider it because we use the Clay OpenAI integration.

Cached inputs save 50% on input tokens for GPT-4o mini, and 90% on GPT-4 when the input stays the same.

For those of us using AI for outbound messaging, this is exactly what we're doing. Running the same prompt structure over and over with different prospect data.

But the prompt needs to stay the same for as long as possible for the inputs to cache.

My old approach was mixing up the prompt by putting variable data (company descriptions, prospect info) in the middle.

The correct way:

Put everything that changes at the very bottom of your prompt.

Keep the instructions, examples, and framework identical at the top.

Only vary the prospect-specific data at the end.

I met with Jeff Barg on the Clay team directly about this and you shouldn't have to change all your integrations. The prompt will cache automatically with how they have the integration set up. 

This might seem obvious to technical people, but this is the ignorance debt you pay when you're not technical.

I'm sorry I made the wrong suggestion for so long.

If you're running AI-powered outbound campaigns and not using cached inputs correctly, you're burning money.

Here's everything I built with Cursor over Thanksgiving weekend. Some things were done in one prompt, some things took me a whole day. 

1. Automated our entire company/contact workflow
One form → Companies to Supabase → Contacts on-demand → Campaign auto-refill at 2am

2. Overhauled analytics

We already had pretty good analytics, but everything would run at midnight, and I wouldn't get snapshots during the day. So, I built this. 

- Snapshots of sending at 12pm, 3pm, 6pm EST
- Warnings if anything looks wrong
- Faster action on issues

3. Built our own email lookup API

One endpoint: first name + last name + domain
Checks our database cache first
Then LeadMagic → Prospeo.io → Icypeas → MillionVerifier guess → No duplicate paying for emails

4. Souped up our Google Maps scraper
- Now finds contacts mid-scrape
- Maps + contact data in one run

5. Automated campaign setup
- Settings done automatically
- Reads copywriting
- Suggests how to structure the Clay table

6. Batch API + cached inputs for scoring
- 50% off a request so we can make the prompts longer with more detail.
- 90% off GPT-5 token costs when you use a cached input on top of the batch API discount. 
- Company scoring at scale

7.  AI Lead Categorization 

- Uses the Smartlead API to pull all unassigned category leads. 
- We use the same prompt with GPT-5 Nano, GPT-40 Mini, and GPT-4.1 Nano. As long as they all agree on what category it automatically uploads the category. 

And now we only have to manually categorize what it doesn't agree on. This was a way harder issue than I thought, which is why I kind of hammered it with the three APIs, and we only have to filter 30-50 replies a day manually.

8. Bulk Contact Enrichment Flow for our biggest senders. 

We have 6 customers sending 50k+ emails per day (1 sending 120k emails per day) that just need a ton of data. 

I'm right this second processing 14M leads to be enriched, qualified, and distributed to the right campaigns. 

Can't wait to see what I can get done this week in Cursor.

Made this video explaining how you can use Cursor to scrape Google Maps and stop messing with custom coding it yourself. Here's the overall guideline to get this done below. 

Why Scrape Google Maps?

LinkedIn isn’t enough: Some businesses, especially local ones, lack a LinkedIn presence.

Google Maps is universal: Almost every legitimate business is listed here.

This method is ideal for local businesses or industries with limited LinkedIn activity.

Tools You'll Need:
Cursor: An AI-powered coding assistant and development environment.

Google Maps Extractor API:
RapidAPI version: Affordable and reliable.
Serper.dev, Outscraper, or similar APIs are also compatible.

CSV of ZIP codes or locations: Optional, but useful for targeted searching.

Clay or CRM: For storing and using your scraped data in email workflows.

Step-by-Step Instructions
1. Get Your API Access Set Up
Visit RapidAPI.com's Google Maps Extractor page (or your preferred API provider).

Subscribe to a suitable plan (e.g., $100/mo for 500,000 requests).

Copy your API key from the provider.

Note: One API request = 1 Google Maps page of results.

2. Open Cursor and Start a Project

Launch Cursor and create a new workspace.
Use natural language prompts like:
"Help me set up a workflow to search Google Maps for businesses in a CSV of ZIP codes. Use the RapidAPI Google Maps Extractor endpoint."
Cursor will write the necessary code for you (this might be Python, JavaScript, etc.).

3. Teach Cursor the API Endpoint
In Cursor, paste in the API documentation from RapidAPI (e.g., the search or locate endpoint).

Show it the input parameters and expected response format.

This improves Cursor’s ability to generate valid code.

4. Build Your Scraping Workflow
Provide prompts like:
"Loop through this list of ZIP codes, perform a Google Maps search for 'restaurants', and save the business name, address, category, and phone number to a CSV."

Modify as needed: e.g., "Go 3–5 pages deep" or "also fetch review data".

Let Cursor adjust the code and test locally within the tool.

5. Once you’ve scraped your data:

Export it into a clean CSV.
Import it into Clay to find owner email and contact information.

Here's how we custom coded a process to make sure all of our cold email campaigns are full of leads and get warnings when they aren't.

For years, it was the same manual process: download CSVs, upload into Clay, repeat.

So we built a system using Cursor, Supabase, Clay, and Railway.

Here are the inputs:

- Campaign ID
- Link to TAM list
- Webhook to Clay

Supabase stores:
- Campaign record
- Job record
- Scraped contacts

Cursor handles the scraping:
- Scrapes all contacts from the provided link
- Stores them in Supabase

The nightly worker (runs on Railway at 2AM ET):

Checks all campaigns + jobs.

Decides how many leads to send to Clay webhook.

Batch size options: 500 / 5,000 / 50,000 per day. really anything that you want

Tracks whether each contact was processed or used before.

Clay staging → processing:

Staging in Clay ensures the contact has never been used for that client.

Approved contacts move to the processing table (cleaning + enrichment).

Auto-delete on the Clay table keeps the system running autonomously.

Smartlead:

Leads flow into active campaigns automatically.

The alerts:

Supabase sends a Slack message when all contacts are processed: "All contacts processed" warning.

This tells us it's time to add more leads or the campaign will run dry.

Optionally for you, we use ClickUp analytics to verify leads are being added.

The outcome: we never have to manually upload CSVs again, and we get proactive warnings before campaigns run out of leads.

How are you currently managing lead intake for your cold email campaigns?

I was just thinking about how far we have come since AI. Here's a list of data points that used to be cutting edge and are now so easy with AI. 

When I first started my agency, one of the things we were asked to do was score a company for any signs of GDPR or SOC II compliance. That used to be a crap ton of google search operators (that each cost .01 per call). 

Now you can use a website markdown like Firecrawl to get what you need and score it with AI.

I had another customer that wanted to know 

1) if a company offered phone support and if so
2) what times were they available. 

Again, a google search nightmare and running tons of if/then formulas to parse out the times that they were available on the phone. 

Now you can easily just ask Claygent to get this information for you from the homepage. 

We also had someone that wanted to take a screenshot of an ecom companies sustainability page and put it in the email. 

Wow so many problems. Finding the page, taking the screenshot, saving the screenshot, finding screenshots that had pop ups and deleting them, etc. 

Now browser automation companies can take care of all that. I'd still find the page with Claygent to get down on that compute time and then have Browserbase navigate to the page, make sure there are no pop ups and then take a screenshot. 

What's possible for sales and marketing teams is just getting crazier and crazier with all the new AI tools.

The Workflow Company

We help B2B startups build and scale a consistent pipeline. We use AI and automation for precision in targeting, hyper-personalization in messaging, and ease in scaling.

Founder and CEO

First Round Fast Track

Mentor

Integromat

I help startups integrate Integromat into their operations to streamline and automate processes

Registered Partner

joerhew.com

I partner with early stage startup founders to bring clarity to their strategy and improve their operations. I help companies prioritize what’s truly important by leveraging data, then help implement a culture of quick iterations and learnings through no code. I like to get my hands dirty and execute, but also act as a strategic thought partner to the founders.

Clients include: Karrot (Daangn Market, Series D, backed by DST), Groove (Seed, backed by Resolute)

Fractional Head of Operations

Ops Hacks

Built a community of ops folks that work at early stage startups

Community Builder

Tyltgo (YC S20)

White-label, same-day delivery service for local merchants across Ontario

Co-founder & COO

Uber

(Sensitive figures redacted)

• Lead a team of 4 operations managers to perform continuous marketplace management of over 100+ municipalities across Canada with focus on growth, reliability, and efficiency
• Launched a series of experiments, features, and processes to consistently exceed targets on growth and costs - increased year-over-year deliveries by x% while reducing courier costs by y% 
• Provided managerial oversight to launch walkers as a delivery mode in Canada, which accounted for c. x% of deliveries in downtown Toronto within y months of launch

Head of Marketplace Operations, Uber Eats Canada

• Managed a cross-functional launch team for 3 months in leading the largest single-day launch in the history of Uber Eats across 30 cities in Canada, which increased national population coverage by c. x percentage points
• Built out Reporting & Performance Management function using SQL and GSheet, providing transparency and enabling accountability on organization’s OKRs for key stakeholders
• Conducted a technical analysis and modeling on x million+ historical transactions to determine the optimal pricing structure and project-managed the pilot of service fees in Canada

Business Operations, Uber Eats Canada

• Collaborated closely with the product team to launch a new carpool product, uberSHARE, the first P2P product since uberX the country's banning of uberX in March 2015
• Launched UberTRIP, an hourly rental product, to target tourists and business travelers and increased weekly city gross bookings by x% without incurring any additional costs
• Successfully overhauled the driver partner incentive program, saving costs equal to c. x% of weekly gross bookings on average with negligible impact on churn and engagement

Rides Country Lead, Uber Korea

Startups

1. One Hive: e-commerce app for international customers interested in purchasing Korean fashion items
• Supported UX and UI design of a mobile e-commerce service to market and sell Korean fashion items to international customers
• As business lead, gained 10K+ users and generated revenue within the first 6 months of launching

2. Foodle: mobile social networking platform with algorithmic recommendations for food enthusiasts 
• As product and business lead, designed UX of the mobile app and led marketing efforts across social and offline channels
• Collaborated with UI designer and front- and back-end programmers to launch product on iOS

Co-Founder / Business Lead

Morgan Stanley

• Built detailed financial valuation Excel models, based on assumptions and hypotheses derived from financial, operational, and economic data, to test feasibility and profitability of envisaged transactions
• Advised large Korean and multinational corporate clients on cross-border M&A transactions and other business development initiatives through providing financial and strategic analyses
• Selected transaction experiences include SK hynix’s US$ 248MM cross-border acquisition of Link_A_Media Devices and MBK Partners’ KRW 1.2Tn (c. US$ 1.0Bn) acquisition of a 30.9% equity stake in Woongjin Coway

Investment Banking Analyst

Citi

• Selected transaction experiences include serving as the Global Coordinator and Joint Bookrunner in Hyundai HCN’s KRW 83.6Bn (c. US$ 74MM) initial public offering and SK Networks’ US$ 700MM minority equity investment in MMX Mineração e Metálicos

We use code and AI to build unique, 1-of-1 GTM engines for scaling startups.

Reach out to chat about AI <> GTM.

We are always testing with new tools but here are some of the mainstays we use on a daily basis:
- Claude Code
- Trigger.dev
- Supabase
- Langfuse
- Railway
- Mastra
- Github
- Attio
- Clay

We are a proud member of the 1% for the Planet and are committed to donating 1% of our annual revenues to help protect the planet.

Joe Rhew

Here's a generally accepted best practice that we've gone full contrarian on:

The conventional wisdom on GTM LinkedIn is that you shouldn't get an LLM to copywrite a full email. Instead, get an LLM to write snippets that you sprinkle throughout the email.

Well, we have been letting Claude Sonnet write full multi-email sequences for a few months now. And it works quite well (see attached image - although not every reply is like that; this particular startup has a very compelling product).

The key difference in how we make it work despite going against the accepted best practices - we've modularized our prompt and compose the full prompt at runtime. We've created a prompt each to share context on the company, product offerings, voice & tone, etc. And we pull these in to create one comprehensive prompt before generating an email sequence.

Here are the main benefits of this approach from my pov:

1. You establish a baseline to measure against. You have an email factory function, which ensures consistency and you can iterate on by updating modules
2. You mitigate drift. The latest contexts live in these modular prompts and will always get pulled into generating emails
3. Iteration tempo increases. You have a hypothesis about a new segment - create a new persona prompt and inject it versus the default prompt to test
4. You create evergreen contexts. You can reuse these modules for inbound, outbound, event invites, product launch emails, etc.

There's a lot more that goes into it - in terms of manual and automated tests and evals to ensure the quality of the output - but that's the gist.

Currently humans (i.e. me) are orchestrating the context composition - depending on the campaign, I am hand-picking which prompt modules should be inserted into the overall prompt. But soon enough we'll be able to delegate this to an agent, who will be able to review each individual recipient and compose the right contexts, which is when things will get really interesting.

Anyone else doing this?

GTM Intelligence Orchestration | The Workflow Company

Well done Drew + the Vercel GTM eng team 👏👏👏

This is definitely the direction that every business should go in with AI - bespoke GTM workflows based on hardcoded business contexts. And trying to SaaS-ify this runs counter to the value prop of full customization tailored to your specific business.

And if you're looking for an agency to do this - we've been doing this for our clients for the past year. Reach out for a brainstorming session.

Original tweet here if you want to learn how Vercel rolled out an AI SDR to remove 90% of the manual labour while keeping key metrics at baseline: https://lnkd.in/gc7a5MUj

Screenshot: 11 days. 121 commits. Complete framework migration.

I just finished migrating our GTM automation platform from Mastra to Claude Agent SDK. Here's why and and what I learned.

---

The Problem with Abstractions

Mastra is a solid framework. But GTM moves fast.

When a prospect starts a new job, you want to reach out *that day*. When a competitor launches a feature, you need a campaign *now*. When your client ships a new product, messaging needs to update *immediately*.

With Mastra, every new campaign meant diving into the codebase. Change the prompts. Update the workflows. Redeploy.

I could have built more tooling around it. But I kept asking: why am I fighting the abstraction instead of just... writing markdown?

---

The Unlock: Markdown as Configuration

With Claude Agent SDK, campaigns are just markdown files.

/campaigns/new-job-signal.md
/campaigns/competitor-response.md
/products/latest-feature.md

Client just raised a round? Update one markdown file.
Launching against a new competitor? Create a campaign brief.
Testing new positioning? Write it in plain English.

No code changes. No deploys. Just context.

The agent reads what it needs, when it needs it. That's it.

---

The Takeaway

Abstractions are great until they slow you down.

For GTM, where signals matter and timing is everything, the ability to iterate in plain English—just markdown files—is the unlock.

Sometimes the best architecture is the simplest one.

Thank you to Jacob Dietle, Jiquan Ngiam, and Yu-Sheng Kuo for your advice - I've picked each of your brains before committing to the migration, and it was definitely the right decision.

We just restructured our prompt architecture and cut LLM costs by 85%.

We've been building GTM automation systems — the kind that research prospects, match them to personas, and generate personalized multi-touch sequences. Each generation prompt runs 10k+ tokens of context: product positioning, messaging frameworks, company narrative, matched personas, lead research.

When you're processing thousands of leads per week, that adds up.

Most of that context is identical across every lead in a batch. Same product. Same messaging. Same company story. Only the persona matching and lead data changes.

So we restructured the prompt to put shared contexts first:

- System prompt ← 100% cache hit
- Shared contexts ← 95%+ cache hit
(products, messaging, etc)
- Resolved contexts ← 60-80% cache hit
(AI-matched personas)
- Lead data ← 0% 
(expected for individual data)

Caching: 90% off repeated tokens

Anthropic's prompt cache gives you a 90% discount on tokens that match a previous request - but only if they appear at the start of your prompt. Reorder your context wrong and you get 0% hits.

Batching: 50% off everything

The Batch API runs async (up to 1hr latency) but halves your token costs across the board. For scheduled outbound campaigns, the latency is irrelevant.

The primer problem

Here's where it got interesting. Batch API processes requests in parallel across multiple workers. Without coordination, each worker warms its own cache independently — killing your hit rate.

So we built a primer mechanism: before submitting the main batch, we fire a single request to warm the cache. Wait for confirmation. Then submit the full batch. Every subsequent request hits the warm cache.

Did reshuffling the context hurt quality?

We backtested ~500 emails — same inputs, same model, old structure vs new. No measurable degradation. The model doesn't care if product context comes before persona context. It all lands in the same attention mechanism.

What this feels like

At 15% of the original cost, we can now be more detailed in our instructions and run and feed deep research contexts. Run more experiments.

It feels like abundance of intelligence - not because the models got cheaper, but because we architected a system to make the right trade-offs (latency for cost) for the use case.

Would recommend caching your prompts if the llm generation task you're running is 1) highly repetitive and 2) requires a lot of input tokens. Relatively straight forward to set up. 

Batching, on the other hand, takes a lot more work to get right but the 50% off discount is well worth the effort if you don't need need the results immediately.

We built a person-level research workflow that injects our company context into every enrichment.

Standard CRM enrichment gives you job title, company size, funding round. 

None of it tells you whether this person matters to your business.

Here's what we did differently:

When someone gets added to an Attio list, our workflow fetches their LinkedIn profile and last 10 posts. But before the LLM analyzes them, we inject our company context:
- Target personas
- Messaging and narrative
- Product capabilities
- Ideal customer profile

The analysis doesn't stop at "who is this person." It extends to "who is this person TO US."

Same person, different client context = completely different fit assessment.

For example, the CEO of a CRM company, might score high for one client's developer tools practice. Score low for another client selling to enterprise sales teams.

The key: we pass a client_id with every enrichment request. The system pulls that client's context from a central database before running the analysis.

Why separate this out?

If I hard code our company context into the prompt, I have to update every enrichment workflow when our personas evolve or product expands.

With a central context database, I update the source once. Every downstream enrichment module inherits the change.

This is the direction every CRM enrichment should go. Not just data about the person. Data about the person in context of your business.

We just spent 3 weeks refactoring our GTM enrichment workflows for agent compatibility. The changes were subtle but critical, and they've shifted how we think about writing reusable code in an agentic setting.

Modularization has always been key to writing elegant code. You create modules that can be easily swapped in and out and shared by multiple workflows.

With agentic capabilities and the ability to expose these modules as tools to the agent to call at their discretion, there's an even stronger call to write modular code.

You create primitives - think of these as building blocks that you often chain together in a workflow - to create deterministic workflows. And not only can you share these primitives across those workflows where you pre-define each step; now you can share them with agents and let them use them however they like to meet the user's objective.

MODULES
[LinkedIn Research] 
[Email Finder]
[CRM Updater]
[Sequence Generator]

WORKFLOWS
[LinkedIn] → [Email] → [CRM] → [Sequence]

AGENTS
[LinkedIn] ↺ [Email] ↺ [CRM] ↺ [Sequence] ↺ [CRM again]

Agents can call these tools in whichever order they deem fit and multiple times, to iterate their way towards an objective.

A lot of our work these days is around thinking about the right way to modularize - how to scope each module and where to draw boundaries, such that modules would work well in both deterministic and agentic workflows.

Here's a real example from last week: We have a module that fetches company records from Attio.
Pre-agents, we'd fetch only the 5-6 fields we knew we needed (company name, domain, employee count, funding stage). Clean, efficient, deterministic.

But when we handed this to an agent trying to personalize outreach, it kept wanting more context. The agent wanted to reference the tech stack, recent funding news, hiring signals - properties we'd deliberately excluded because we didn't need them in our scripted workflows.

The fix wasn't adding those specific fields, but rahter returning the full record and letting the agent decide what's relevant. This made the module "fatter" but infinitely more reusable in agentic contexts.

In practice, this means:
- Our Attio retrieval tasks now return full objects, not filtered subsets
- Our Supabase queries use SELECT * more liberally than we used to
- Our Claude Code workflows document all available fields, even "optional" ones

There's an irony here. Good modular code for agents looks "sloppier" than good modular code for workflows. You're returning more data, exposing more parameters, being less opinionated about usage. It feels wasteful - until you see an agent solve a problem you didn't anticipate by combining modules in a way you never would have scripted.

🧠 Context: I received a request for case studies from a potential client
💪 Me: Prompt my Claude Code agent - 
"Look at the most recent email from [Prospect - let's call them John] asking for case studies.

I've sent something similar to:

[List of prospects/clients]

Then look up the meeting transcript with [John] from yesterday to get more context on what he does [in his current role]. I mentioned [specific contexts we should highlight in the case studies], so think about that + other contexts shared during the convo on how to structure the case studies.

Review all of this and let me know if you have any questions before drafting the email response to [John].

Feel free to use multiple subagents/tasks to do this."

🤖 Agent:
1. Parallelized searches for:
A. previous emails to prospects
B. previous emails that mention case study
C. Email from John
D. Transcript from meeting with John

2. Synthesized the request and the available context

3. Drafted email that was 80% of the way there and I dictated my feedback to get it to 100%

--

A few top of mind callouts after doing this:

1. Funnelling all of your contexts into a unified context layer is key to unlocking agentic capabilities. I save my transcripts in a repo, readily accessible to the agent; and the agent has API access to search, read and draft emails (but not send - human in the loop!)

2. Getting the agent to ask you questions to fill in the gaps / find edge cases is a simple hack to getting better output that's more in line with how you've thought about it

3. Given how many times I've had to send case studies via email, it's probably time for me to put them up on our website. Claude now has rich context on this based on all the emails, so this will be a relatively straightforward process <- Involving Claude Code in every task creates a compounding effect where subsequent tasks get easier, thanks to having more context

4. *Most important* Claude Code isn't going to replace you, but someone using Claude Code will. 

I've been thinking about creating a series of walkthroughs on using Claude Code for GTM - would you watch it or is working in the CLI an instant turnoff?

taplio

The Battle

🦾Eric Nowoslawski

Joe Rhew

Followers

Posts

Engagements

Likes

Performance Score🦾Eric Nowoslawski's LinkedIn Performance

Performance ScoreJoe Rhew's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

🦾Eric Nowoslawski

Joe Rhew

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}🦾Eric Nowoslawski's LinkedIn Performance

Performance ScoreJoe Rhew's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance Score🦾Eric Nowoslawski's LinkedIn Performance