This blog is where we will publish practical notes on building and operating that system.

What We Will Cover​

• GTM agents for account sourcing, qualification, research, buyer discovery, and outreach preparation.
• Reusable skills for ICPs, personas, eligibility rules, fit scoring, sender context, and sequences.
• Business data patterns that help agents reason about companies, people, contacts, and market events.
• Workflow design for keeping humans in control while agents perform repeatable GTM work.

How To Use This Blog​

Articles are written as Markdown files in this repository and published as static HTML. That keeps each article readable by browsers, search crawlers, and AI discovery tools.

For machine-readable context, start with:

Canonical Description​

Truebase is an AI workspace for GTM teams. It helps teams define reusable GTM skills, run agent workflows, use business data, and review outputs before they move into sales and marketing systems.

For a long time, the framing was simple. Salesforce was the enterprise CRM. HubSpot was the SMB and mid-market CRM.

That framing is starting to feel outdated.

The new battle is not just CRM vs CRM. It is the battle to become the operating layer for GTM teams.

That means the system that connects customer data, account signals, buyer context, AI agents, and human review into one repeatable motion.

The latest public numbers make that shift easier to see.

Salesforce is now reporting its business around agentic apps and data infrastructure. HubSpot is describing itself as an agentic customer platform, not just an easier CRM. Both companies are telling the market that the next product surface is not the database of record. It is the execution layer on top of the record.

What Changed In The Last Few Weeks​

Salesforce's latest quarter is a filing-level signal, not just a keynote narrative.

In its Q1 FY27 results release filed with the SEC, Salesforce reported:

Marc Benioff framed the quarter as "record revenue, record deals, and cash flow." He also called agentic AI Salesforce's biggest growth opportunity.

The more important detail is in Salesforce's Q1 FY27 Form 10-Q. Salesforce revised its disaggregated subscription revenue categories into two primary groups: Agentforce Apps, and Data 360, Headless Platform, and Other. That is not just a naming change. It says Salesforce wants investors, customers, and operators to understand its architecture as apps plus data/platform.

HubSpot is making a parallel move from the other direction.

In its Q1 2026 results, HubSpot reported:

HubSpot's 2025 Form 10-K describes the company as an agentic customer platform with three layers: AI-powered agents and engagement hubs, Smart CRM, and a connected ecosystem. It also says Breeze powers the customer platform, including Smart CRM, engagement hubs, and the ecosystem.

On HubSpot's Q1 2026 earnings call, Yamini Rangan made the strategic point directly:

"AI with the right context produces outcomes."

She also said HubSpot is "not building AI features on top of CRM."

That distinction matters. HubSpot is arguing that the CRM is the context layer, while agents operate on top of it and through it.

The Old CRM Boundary Is Blurring​

The old segmentation was mostly about company size and implementation complexity.

Salesforce was the default choice for large organizations that needed deep customization, complex permissioning, partner ecosystems, and heavy RevOps control.

HubSpot was the default choice for teams that wanted speed, usability, marketing automation, and a system the business could operate without turning every change into a platform project.

That distinction still matters, but it no longer explains the market.

Modern GTM work is not neatly contained inside a CRM record. Account selection, buyer discovery, enrichment, qualification, routing, research, outbound prep, lifecycle triggers, and expansion signals all depend on data and workflows that live across many systems.

The system that wins is not just the system that stores the account. It is the system that helps the team decide what should happen next.

The Real Battle Is Execution​

Every GTM team is trying to answer the same operational questions:

• Which companies should we care about?
• Which people matter inside those companies?
• What changed recently that makes outreach relevant?
• What should a rep, marketer, or agent do next?
• Which steps need human review before anything reaches a customer?

Those questions are not solved by a CRM field alone. They require a working layer that can combine company data, people data, customer data, market events, business rules, and workflow context.

That is why both companies are moving toward agentic execution.

Salesforce is pointing to volume and infrastructure: trillions of records, API calls, tokens, and a product architecture reorganized around Agentforce and Data 360.

HubSpot is pointing to business usability: agent workflows that sit inside the customer platform, pull from CRM context, and help sales, marketing, and support teams execute without rebuilding the whole stack.

The next phase asks both companies a harder question: can they become the place where GTM work is orchestrated, not just recorded?

AI Makes the Gap More Obvious​

AI does not remove the need for GTM systems. It exposes which systems have usable context.

An AI workflow is only as good as the data, rules, permissions, and review steps around it. If a team cannot clearly define its ICP, eligibility rules, fit criteria, sender context, sequence strategy, and approval boundaries, then the agent has to guess.

That is risky.

This is why the next GTM platform will need more than prompts attached to CRM data. It will need reusable business logic that agents can execute consistently.

For example:

• An ICP should be a reusable skill, not a one-off prompt buried in a workflow.
• Qualification should use explicit fit and eligibility rules, not vague scoring text.
• Research should cite the company, person, and event data it used.
• Outreach prep should separate agent-generated drafts from approved customer-facing actions.
• Human review should be a first-class part of the workflow, not an afterthought.

Once AI enters the GTM stack, the CRM becomes one part of a larger operating model.

HubSpot's New Activity Is Very GTM-Specific​

HubSpot's Spring 2026 launches are not just generic AI assistants. They are pointed at common GTM execution gaps.

On its Spring 2026 Spotlight page, HubSpot says Prospecting Agent now handles the entire prospecting flow using CRM data, buying signals such as funding rounds, and product details. It also says Smart Deal Progression uses emails, notes, deal activity, pipeline definitions, and forecast logic to suggest updates after calls.

The earnings transcript adds adoption data:

This is why HubSpot is more interesting than "SMB CRM" suggests. Its wedge is operational: make the CRM update itself, make prospecting run from live context, make support resolution outcome-priced, and let teams use AI without becoming platform engineers.

What Salesforce Has to Prove​

Salesforce has to prove that enterprise power can become operationally usable.

Large GTM teams have complex data models, governance requirements, sales motions, partner channels, and reporting needs. Salesforce is strong in that environment. The challenge is that complexity can slow down experimentation and make business teams dependent on specialized platform work for every operational change.

In an AI-assisted GTM world, that friction becomes more visible.

Teams will want to change ICP definitions, test new account signals, add new qualification rules, adjust research logic, and route agent outputs into review flows quickly. If those changes require too much platform ceremony, the workflow moves elsewhere.

Salesforce does not need to become simple in the same way HubSpot is simple. It needs to make complex GTM operations easier to express, govern, and run.

The company clearly has the data and enterprise footprint. Its Q1 release says more than 50% of Agentforce and Data 360 bookings came from existing customers. That is a strong expansion signal. But it also raises the product challenge: if the opportunity is mostly inside existing Salesforce estates, Salesforce has to make agentic workflows feel like leverage, not another layer of admin work.

What HubSpot Has to Prove​

HubSpot has to prove that simplicity can scale into more complex GTM operations.

Its strength has always been that teams can move quickly. That matters even more when GTM motions change quickly and AI workflows make experimentation cheaper.

But as HubSpot moves into larger and more sophisticated teams, the bar rises. Those teams need stronger data models, governance, permissioning, lifecycle control, integration depth, and workflow reliability.

The risk for HubSpot is not that it is too simple. The risk is losing the simplicity that made it valuable while trying to handle the complexity that larger teams require.

The company that wins this layer will make advanced GTM work feel operable without making it feel fragile.

HubSpot's Q1 numbers show real momentum, but also a different scale than Salesforce. HubSpot guided to roughly $3.7 billion in full-year 2026 revenue. Salesforce guided to $45.9 billion to $46.2 billion in FY27 revenue. HubSpot is not winning by being bigger. It has to win by making agentic GTM work easier to adopt and easier to trust.

The Platform That Matters​

The winning platform will help GTM teams turn strategy into repeatable execution.

That means the important product surface is not only the CRM object model. It is the layer where teams define:

• ICPs and market segments.
• Eligibility and fit rules.
• Account and contact research workflows.
• Buyer and persona logic.
• Event and intent interpretation.
• Sequence and sender context.
• Review, approval, and handoff steps.

This is also where AI agents become useful. They should not be generic chat boxes pointed at CRM data. They should run bounded workflows with explicit skills, clear context, and structured outputs.

The public data points to the same conclusion from both sides.

Salesforce is telling the market that agentic apps and data infrastructure are now central enough to reshape revenue reporting. HubSpot is telling the market that agents, Smart CRM, and growth context are now central enough to redefine the company category.

The future GTM stack will still need CRM systems. But the strategic question is shifting.

It is no longer, "Which CRM stores our GTM data?"

It is, "Which system helps us turn GTM strategy into governed, repeatable action?"

That is why HubSpot vs Salesforce is becoming such an interesting battle.

It may be the person who can manage the most context across disciplines.

For years, building something required a full cross-functional team.

A product manager to define the problem.

A designer to shape the experience.

An engineer to build it.

A QA person to test it.

A marketer to position it.

A growth person to distribute it.

A leader to keep everyone aligned.

That structure made sense because execution was fragmented.

Every function had its own tools, language, backlog, and handoff.

But AI agents are starting to collapse some of those handoffs.

One person can now move from idea to prototype to copy to design feedback to testing to launch much faster than before.

The Bottleneck Moves To Context​

The old bottleneck was often production capacity.

Could we write the copy? Could we build the prototype? Could we analyze the spreadsheet? Could we summarize the research? Could we generate the first draft?

AI reduces the cost of many of those first passes.

That does not make expertise irrelevant. It changes where expertise shows up.

The bottleneck moves upstream to context:

• What problem are we solving?
• Who is it for?
• What constraints matter?
• What does good look like?
• Which data should the agent use?
• Which assumptions should it not make?
• Which outputs need review?
• Which system should receive the final result?

The person who can answer those questions across disciplines becomes much more valuable.

Specialists Still Matter​

This is not an argument against specialists.

Deep expertise still matters for judgment, taste, correctness, safety, and strategy. A generalist with AI cannot replace the judgment of a strong engineer, designer, lawyer, security expert, researcher, or operator in high-stakes work.

But the shape of collaboration changes.

Specialists may spend less time producing every first draft and more time setting standards, reviewing edge cases, encoding judgment into reusable workflows, and improving the systems that agents use.

The New Operator Looks Different​

The valuable AI-era operator is not just "good at prompts."

They are good at translating across domains.

They can talk to sales and understand buyer context. They can talk to product and understand constraints. They can talk to marketing and understand positioning. They can talk to engineering and understand what can be automated safely. They can talk to leadership and understand the decision that actually matters.

Then they can package that context into an agent workflow.

That is the new leverage.

What This Means For GTM Teams​

GTM work has always been cross-functional.

An account strategy touches sales, marketing, data, product, customer success, finance, legal, and leadership. The hard part is not only producing the artifact. It is maintaining the context across all of those functions.

AI agents make this more obvious.

The team does not just need someone who can ask for an email draft. It needs someone who can define:

• The ICP.
• The buyer persona.
• The fit and eligibility rules.
• The account signals.
• The source data.
• The sender context.
• The sequence strategy.
• The approval boundary.

That person may become one of the highest-leverage operators on the team.

The Skill Is Context Management​

The AI era rewards people who can hold a complex situation in their head, break it into reusable logic, and guide agents without losing the nuance.

That is not shallow generalism. It is context management.

The deepest specialist still matters.

But the person who can connect the most useful context across disciplines may become the person who moves the fastest.

Mail, Slack, docs, spreadsheets, passwords, bookmarks, writing tools, all of it lived in the browser.

It made my laptop feel almost disposable.

My real work environment was not the machine. It was the browser state.

That model was incredibly powerful in the SaaS era.

The Browser Won On Access​

The browser became the universal interface because most software was built for humans to operate directly.

The loop was simple:

1. Read.
2. Click.
3. Type.
4. Move to the next tab.

That interaction model worked because SaaS made software accessible from anywhere. The operating system mattered less. The browser session mattered more.

For many knowledge workers, the browser became the actual workspace.

The numbers match that feeling.

Gartner forecast worldwide public cloud end-user spending would reach $723.4 billion in 2025, with SaaS alone projected at $299.1 billion.

Okta's Businesses at Work 2024 report found the average company deployed 93 apps, up 4% year over year. U.S. companies averaged 105 apps. Dutch companies averaged 108.

That is the browser-native era in one statistic: more work, more apps, more tabs, all made reachable through the same universal access layer.

Agentic AI Changes The Interface Question​

I have been noticing something different with agentic AI.

The more software shifts from helping you work to doing work on your behalf, the more the interface starts to matter in a different way.

Not just where you access the product.

Where the product can act.

That changes the game.

Controlling a browser tab is one thing. Controlling files, desktop apps, system permissions, local state, long-running jobs, and background workflows is another.

Agents need a deeper control layer than the browser was designed to provide.

That is already visible in how the frontier products are described.

OpenAI's Computer-Using Agent research framed the model as using a "universal interface" for digital work: screen, mouse, and keyboard. It reported 38.1% success on OSWorld for full computer-use tasks, 58.1% on WebArena, and 87% on WebVoyager.

Anthropic described Claude's computer use as teaching the model general computer skills so it can use standard tools and software programs designed for people. Its own caveat matters too: the capability is still imperfect, and low-risk tasks are the right place to start.

That combination is the point. The industry is not only trying to make chat better. It is trying to make agents operate software.

From Access To Action​

The browser won the last era on access.

Native may win the next one on action.

That does not mean the browser goes away. It means the browser may stop being the only center of gravity for work.

In the SaaS era, the key question was:

"Can I access this product from anywhere?"

In the agentic era, the key question becomes:

"Can this agent safely act across the systems where work actually happens?"

Those are different product requirements.

McKinsey's State of AI survey shows why this matters now. Seventy-eight percent of respondents said their organizations used AI in at least one business function, up from 55% a year earlier. Seventy-one percent said their organizations regularly used gen AI in at least one business function.

The functions showing up are not just writing surfaces. McKinsey points to IT, marketing and sales, service operations, product and service development, software engineering, and knowledge management. Those are workflow surfaces.

Once AI moves into workflow surfaces, access is not enough. The product has to know what action is allowed, what context is trusted, which state should change, and where human review is required.

Remote Machines May Become More Valuable​

If this shift continues, rented remote machines may become much more valuable.

Not just as infrastructure, but as parallel operating environments for agents working on your behalf.

An agent may need a place where it can:

• Keep authenticated sessions alive.
• Work across browser tabs and native apps.
• Read and write files.
• Use local tools.
• Run long jobs.
• Preserve state between steps.
• Operate under scoped permissions.
• Hand work back for human review.

That starts to look less like a chatbot and more like an operating environment.

This is why remote execution may become strategically important. A useful agent needs more than a prompt box. It may need its own durable workspace: browser sessions, files, tools, logs, secrets, permissions, queues, and review checkpoints.

If the work takes minutes, hours, or days, the environment matters. If the agent needs to pause for approval and resume later, the environment matters even more.

What This Means For GTM Work​

GTM work already spans too many surfaces:

• CRM records.
• Email threads.
• Call notes.
• Spreadsheets.
• Sales engagement tools.
• Data providers.
• Research tabs.
• Internal docs.
• Approval workflows.

The browser can expose many of those systems. But an agent that actually completes the work may need to move across them, preserve context, run multi-step tasks, and return structured output.

That requires more than a browser extension. It requires a governed action layer.

The same pattern shows up in GTM data. A prospecting workflow may need to read CRM fields, inspect a company website, check funding news, compare against ICP rules, find contacts, draft outreach, and wait for approval before anything is sent.

Each step is small. The workflow is not.

The hard part is not opening the tools. The hard part is carrying state across tools without losing the business rules that make the action safe.

The Next Interface Shift​

We spent the last era making software browser native.

This next era may be about making software agent native.

That means designing products so agents can understand the state, use the right tools, respect permissions, produce reviewable outputs, and complete work safely.

The browser is still important. But the center of gravity may shift from where humans click to where agents can act.

A few years ago, public internet data was the prize. It was human, messy, massive, and still largely untapped.

Now a growing share of the web is AI-generated. The public layer is getting flatter, and in many places, less original.

At the same time, our behavior is changing. We are feeding these systems everything: docs, chats, images, spreadsheets, notes, memory, and context.

More context in, better output out.

That shift is already visible in the data. Microsoft's 2024 Work Trend Index, based on 31,000 people across 31 countries, found that 75% of global knowledge workers use AI at work and 78% of AI users bring their own AI tools to work. McKinsey's 2025 State of AI survey found that 88% of organizations report regular AI use in at least one business function, but only about one-third say they have begun scaling AI programs.

The Web Is Becoming Less Differentiated​

The public web still matters. It contains facts, signals, writing, code, product pages, research, forums, company updates, and market context.

But public data has a problem: everyone can reach it.

It also has a supply problem. Epoch AI estimates the effective stock of quality-adjusted, human-generated public text for AI training at roughly 300 trillion tokens, with current trends pointing to full use of that stock sometime between 2026 and 2032.

And the public layer is getting noisier. In a 2024 Nature paper on model collapse, researchers found that indiscriminate training on model-generated content can cause "irreversible defects" in future models. Their conclusion was blunt: as LLM-generated content spreads across the web, data about genuine human interactions becomes more valuable.

If every major model can learn from roughly the same public layer, the base model eventually becomes less differentiated by that layer alone. The next advantage moves to the data that is not broadly available, not broadly licensed, and not automatically included in the training mix.

That is where private context becomes strategic.

Users Are Voluntarily Supplying Better Context​

The trade often feels small in the moment.

More context in, better output out.

You upload the spreadsheet because the analysis gets better. You paste the email thread because the reply gets more precise. You connect the document repository because the assistant can answer with more relevance. You share the meeting notes because the next step becomes clearer.

Zoom out, and something bigger is happening.

The next important AI input stream may not come from crawling the web. It may come from users voluntarily pouring private context into consumer and business AI tools.

Google made the consumer version explicit in January 2026 with Personal Intelligence, a Gemini feature that can connect Gmail, Photos, YouTube, and Search. The framing was simple: the best assistants do not just know the world; they know you. Google also separated reference from training, saying private Gmail and Photos content can be used to deliver a reply without directly training the model on that content.

That makes this more than a privacy topic. It is a strategic topic.

The Moat May Be The Policy Layer​

If the major models keep converging at the base layer, the real differentiation starts to move elsewhere:

• Who gets access to the best private data?
• Under what defaults?
• On which plans?
• With what user consent?
• With what enterprise controls?
• With what retention and deletion rules?
• With what separation between inference context and model training?

That is where the next moat may form.

The winning AI platforms will not just have the best model. They will have the best permissioned context graph.

This is why privacy and retention language is starting to read like product strategy. OpenAI tells business customers, "We don't train our models on your organization's data by default". Google Workspace says customer data is not used to train or fine-tune generative AI models without prior customer permission or instruction. These are not just legal assurances. They define who can contribute context, when it can be reused, and whether the platform can compound value from private workflows.

Why This Matters For GTM Teams​

Go-to-market teams live inside private context.

The public internet can tell you what a company says about itself. Private GTM data tells you what has happened with that company:

• Which reps have spoken to which buyers.
• Which objections appeared in prior calls.
• Which campaigns touched the account.
• Which product features were mentioned.
• Which competitors came up.
• Which legal, security, budget, or timing constraints matter.
• Which internal stakeholders need to approve the next step.

That data is far more useful than a generic company summary.

It is also far more sensitive.

Salesforce's 2024 State of Sales research shows why this matters operationally. Reps reported spending 70% of their time on non-selling work. B2B buyers were 86% more likely to purchase when companies understood their goals, but 59% said reps do not take the time to understand their unique challenges and objectives. Salesforce also found that only 35% of sales professionals completely trust the accuracy of their organization's data.

That is the GTM version of the private-context problem. The useful signal is not just an account name or a scraped company profile. It is the messy operating memory around the account: the call history, objection trail, buying committee, CRM activity, campaign touches, legal requirements, and rep judgment that never shows up cleanly on the public web.

The Important Question Is Not Just Model Quality​

The important question is not only which model is strongest.

It is which system can safely use the right private context at the right time, for the right purpose, under the right controls.

For GTM teams, that means an AI workspace has to distinguish between:

• Context used to complete one task.
• Context saved as workspace memory.
• Context exposed to a specific agent.
• Context approved for downstream CRM or sales system actions.
• Context that should never leave the workflow.

Those boundaries are not implementation details. They are product strategy.

The moat may not just be the model. It may be the policy layer that decides what the model gets to use, what it gets to remember, and what it is allowed to do.

Good prompting starts with giving the full picture.

That usually means pulling together the emails, docs, chats, meeting notes, and everything else that matters.

In the process, you are forced to slow down, reflect, and reconstruct the situation.

That alone is powerful.

The Prompt Is Not The First Step​

The visible work is the prompt. The useful work usually happens before it.

When someone asks an AI system to write the next email, summarize an account, help with a strategy memo, or prepare a decision, the model needs more than the immediate request. It needs the background: what happened, who is involved, what changed, what constraints exist, what has already been tried, and what outcome matters.

That preparation changes the user as much as it changes the model output.

Before AI, many of us would jump straight into the email, memo, deck, or decision. Now the better pattern is to first build the context window.

Context Building Feels Like Journaling​

There is something quietly useful about gathering scattered inputs and turning them into a coherent narrative.

You are not just feeding the model. You are organizing your own thinking.

That is why this often feels like journaling. You take a messy situation and make it legible. You capture the facts, the assumptions, the emotional charge, the missing details, and the next decision.

There is a structured version of this in cognitive behavioral therapy: capture the situation, the thoughts, the evidence, and then reframe.

The same pattern shows up in writing. A good screenwriter does not start with random dialogue. They build the character, the backstory, the motivation, and the scene. Only then does the story make sense. Only then does the next line feel right.

Knowledge work is starting to work the same way.

The Context Window Becomes A Work Surface​

The context window is more than a technical limit. It is becoming a work surface.

For a GTM team, context might include:

• The account history.
• The current opportunity stage.
• Relevant emails and call notes.
• The buyer's role and likely priorities.
• The company's recent funding, hiring, product launches, or market events.
• The ICP, fit, eligibility, and persona rules that should guide the work.
• The sender's style, constraints, and relationship to the account.

Once that context is assembled, the AI output becomes less generic. More importantly, the human has a better mental model of the situation.

One Chat Per Topic Is Often Better​

I have found it useful to keep one chat per topic, with all the context, and keep it alive for the life of the deal, project, or opportunity.

When a chat gets too long, it can lose the thread and miss important detail from earlier context. The better pattern is to treat long conversations as working folders, not infinite streams. Keep the important context together, but periodically compress it into a clean brief.

That creates a reusable asset:

• What is true?
• What changed?
• What still needs to be decided?
• What does the agent need to know before it acts?

This is where AI chat becomes more than a writing assistant. It becomes a way to maintain a living situation model.

The Habit Matters Even Without AI​

AI or not, the discipline of gathering context first, then thinking, then writing is a very good habit.

It slows down reactive work. It makes hidden assumptions visible. It gives teams a shared version of the situation. It reduces the chance that the next action is based on a half-remembered thread or a stale CRM field.

For GTM teams, that matters because the expensive mistakes are usually context mistakes:

• Reaching out with the wrong premise.
• Treating a qualified account like a generic lead.
• Missing a buying signal.
• Forgetting the prior objection.
• Asking the rep or agent to act before the situation is understood.

The AI era may make context building more important, not less.

The best operators will not just write better prompts. They will build better context.