Welcome to the first edition of Juniply Field Notes.

Go-to-market is moving quickly. Every week brings another model, agent, workflow, or company promising to transform how we sell and market.

I’m starting this newsletter to help make sense of it—but I don’t want it to become another list of AI headlines.

I want to test the tools, share what people are building, and show what is actually possible.

Because my current take on AI agents is simple:

Agents can accelerate growth, but a tool alone won’t create the 100× outcome everyone keeps promising.

The problem matters.

The idea matters.

The tools matter.

The workflow connecting them matters most.

That’s what I’ll explore through Juniply Field Notes.

I used to book demos with almost every interesting YC company I encountered.

I wanted to understand what founders were building, which problems they were trying to solve, and where AI was going next.

I still approach new tools with that curiosity. But the more tools I test, the less convinced I am that any single product will create the transformational result we’re looking for.

One model may be better at writing code. Another may be better at design. A third may connect the resulting system to your CRM.

None of those capabilities means much until they are applied to the right problem.

That’s when an agent stops being a demo and starts becoming leverage.

OpenAI recently released the GPT-5.6 model family, including Sol, Terra, and Luna.

I’ve been using GPT-5.6 for coding and image generation, and it has been genuinely impressive. I’ve found it effective even compared with larger competing models.

But not every workflow needs the most capable—or most expensive—model.

There are still countless straightforward tasks worth automating. When they run repeatedly, choosing a model becomes a cost and speed conversation, not simply an intelligence benchmark.

Use powerful models where ambiguity and judgment matter. Use faster, less expensive models for the repeatable work surrounding them.

The best model increasingly depends on the job.

OpenAI Presence helps companies deploy voice and chat agents that can use company knowledge, interact with systems, take approved actions, and escalate to humans.

I’ve built voice-agent experiments with ElevenLabs, so I’m excited by the possibilities.

The external use cases are obvious: support, qualification, onboarding, and product education. The internal opportunities may be just as interesting.

But there’s a difficult question underneath all of this:

How does an agent know which company information is current, outdated, or simply wrong?

People are already bad at maintaining documentation. As companies ship faster, the knowledge layer becomes even harder to manage.

Building the voice is getting easier.

Building a trustworthy source of truth may become the more valuable problem.

HubSpot recently introduced Agent Hub and Agent Builder, giving teams one place to build, manage, and review agents operating across their go-to-market motion.

I already build agents that pull data out of HubSpot, so I don’t currently rely on HubSpot for most of my agent workflows or dashboarding.

However, one of the hardest questions after introducing an agent is still:

So what?

Did it improve pipeline?

Did it reduce response time?

Did it create revenue—or merely more activity?

Individual agent features can be copied. HubSpot’s stronger advantage may be connecting what an agent does to the customer and revenue data already inside the CRM.

Clay’s new Account Research Agents provide continuously updated intelligence across a company’s book of business.

I’ve used Clay for years. Although I’ve moved much of my work away from it, I still think it is an excellent product.

For a large enterprise account with multiple representatives and stakeholders, continuously monitoring calls, emails, signals, and account changes could be valuable.

For smaller companies, I’m less convinced that “always on” is necessary.

A thoughtful daily or weekly update may be enough.

The real opportunity isn’t constant information. It’s identifying the few changes that require action.

If an agent monitors everything but cannot explain what affects pipeline health, it has created another feed someone needs to manage.

Salesforce has signed an agreement to acquire Fin, formerly Intercom.

It’s another sign that the large CRM platforms see AI agents as a major platform shift.

As traditional SaaS companies face pressure, I expect more acquisitions and consolidation. CRM companies already have the customer data and distribution. Emerging AI companies have speed, specialized products, and new interfaces.

The next stage may not be about who builds the best standalone agent.

It may be about who owns the environment where agents work—and where their results are measured.

I’m particularly interested in what HubSpot does next.

One of my most useful discoveries has been that I don’t need to choose one AI tool for an entire project.

I’ve been using Claude for code writing, architecture, and longer implementation sessions.

I’ve been using Codex for design direction, image generation, and work involving connected files.

One strangely specific example: Codex can update an existing Google Doc directly, while some of my other workflows keep trying to create another file. That sounds minor until you have five versions of the same document and no idea which one is current.

I’ve also been saving image-generation credits by using Codex for much of the visual work.

My current workflow looks something like this:

This isn’t a permanent ranking. The tools will keep changing.

The broader lesson is more important:

You don’t need one AI tool to do everything. You need a system that uses each tool where it creates the most leverage.

Five rounds. Four representatives. A few seconds to make each pick.

This week, I built Waiver Wire: a live, NFL-style draft that lets sales representatives choose inbound leads during a sales meeting.

Traditional territory planning relies on static rules:

Those rules are useful once a company has a mature motion. But while a team is scaling and still learning its ICP, they can be rigid—and honestly, a little boring.

With Waiver Wire, a sales leader starts a draft during an all-hands meeting or sales stand-up. Representatives review available leads, see company information and fit scores, and claim the accounts they want before the clock expires.

An AI announcer provides live commentary as the picks come off the board.

At the end, each representative receives a scouting sheet containing their draft grade, total potential value, strongest picks, and assigned accounts.

Sales leaders can export the results or push the new ownership assignments into a CRM such as HubSpot, Salesforce, or Act-On.

I’ve run plenty of sales meetings throughout my career. The operational work needs to happen, but it doesn’t have to feel like a chore.

For the next few months, I want to build more tools that bring some whimsy back into work.

Just because it’s work doesn’t mean it can’t be fun.

Watch the Waiver Wire demo

Explore the public repository

You have clients to visit, conferences to attend, and a limited travel budget.

Who should you see? Which events are worth attending? Should you drive, take a train, or fly?

Circuit applies the traveling salesperson problem to go-to-market.

It combines potential events with existing customer accounts and creates a travel plan based on:

A company could use a crawler to discover regional conferences, association meetings, and university demo days. Circuit then ranks those opportunities against the company’s ICP and combines them with customer visits.

The purpose isn’t merely to find the shortest route.

It is to find the trip that creates the best return.

We spend a lot of time discussing how AI can remove human interaction.

I’m interested in how it can help create better human interactions.

I’ve worked at startups for most of my career, and major conferences can be incredibly expensive. Smaller regional events can provide concentrated access to the right audience at a fraction of the cost—but they are much harder to discover and compare.

If you’re already traveling through a region, which customers could you meet along the way? Which accounts are at risk? Which relationships would benefit from an in-person conversation?

AI can automate another email.

What it can’t replace is your pretty face showing up in person.

Watch the Circuit demo

Explore the public repository

I recently spoke with Dean from Sphinx about what it means to be AI-native in go-to-market.

One of the most useful ideas from our conversation was that you shouldn’t begin by automating a process.

First, perform it manually from end to end.

Understand where the work gets stuck. Identify what requires judgment. Then automate the small, repeatable parts that clearly deserve it.

Otherwise, it is easy to spend a week building a beautiful workflow only to discover that the process is changing—or wasn’t worth automating.

We also discussed what separates someone effective with AI from someone who simply has access to the tools.

It came down to two qualities:

AI can gather context and execute work. The human still needs to decide what deserves attention, recognize when the system is drifting, and determine whether the result meets the quality bar.

We may not be replacing every individual contributor.

We may be turning more individual contributors into managers of agents.

Watch the full conversation

I want this newsletter to become participatory—not simply something you read.

The idea is to invite companies to submit real go-to-market problems. Builders can then create workflows, agents, or lightweight tools that attempt to solve them.

For the first challenge:

Spend no more than 30 minutes documenting:

You don’t need to build anything yet.

Reply with your workflow or send me a screenshot of your notes. I’ll choose one submission as a future Juniply build challenge.

For larger challenges, I’d love to bring finalists together for a live, Dragon’s Den-style demonstration.

Have you built a useful workflow, agent, prompt system, or strange AI experiment?

Send it to me.

It doesn’t need to be polished. It does need to solve a recognizable problem or teach us something useful.

Future editions will feature one community build, including how it works, who should try it, its limitations, and what its creator learned.

Before automating your next workflow, map it manually:

Trigger → Inputs → Decisions → Actions → Result

Circle every step requiring judgment.

Underline every purely repetitive step.

Automate one underlined step—not the entire process.

Then compare the time, quality, or result before and after.

That’s a much better starting point than asking an agent to “automate my GTM.”

Juniply Field Notes will be published bi-weekly.

I’m testing tools, building strange ideas, and learning what works in public.

Explore Juniply

Follow the 52 Weeks of Building challenge on YouTube

If you know someone who is also trying to understand where modern go-to-market is headed, forward them this edition.

The 100× probably won’t come from one tool.

It will come from learning how the pieces fit together.

Until next time,

From Brendan