← All essays

Your tech stack has 40 tools. AI should kill 15.

A line-by-line audit of the GTM stack and what one agent quietly made redundant.

Your GTM stack has somewhere around 40 tools, and a capable agent plus a couple of primitives can quietly retire close to 15 of them. The cut is not about being cheap. It is about deleting the integration tax, the seat sprawl, and the shadow renewals that pile up when every job gets its own logo.

How the stack got to 40

Nobody bought 40 tools on purpose. The stack grew the way most do: a rep found a clever Chrome extension, a marketer signed a list-building contract, RevOps bolted on a dashboard, someone added a note-taker because the last one logged out mid-call. Each decision was rational. The sum is not.

The damage is rarely the line item. A single-purpose tool at a few hundred dollars a month feels harmless. The real cost is the connective tissue. Every tool needs an integration, an owner, a renewal date, a security review, and a place in someone's daily routine. Multiply that by 40 and you are running a small systems-integration shop inside your sales org without meaning to.

An agent changes the math because most of these tools are not products. They are jobs wrapped in a UI. Once one system can do the job on demand, the wrapper stops earning its seat.

The audit: walk it line by line

Open the billing export. Put every tool in one column and the actual job it does in the next. Be specific. Not "enrichment platform" but "finds a title and a verified email for a contact." Then ask the only question that matters: does the agent already do this job, or could it with access to the right data source?

Here is where the cuts usually land.

• Standalone enrichment. A tool whose whole purpose is appending a title, a headcount, or an email to a record. An agent with access to your data layer and the open web does this inline, in context, at the moment you need it. The standalone subscription becomes redundant.
• Sequencing add-ons. The bolt-on that schedules a five-touch cadence on top of your existing email system. This is the work an AI SDR now owns end to end (covered in its own essay), so the add-on is paying for a feature you bought twice.
• Scheduling tools. The calendar-link widget. Useful, low cost, but increasingly a default feature of the system of record or the agent itself. Worth a hard look at renewal.
• Note-takers and recap bots. The meeting transcriber that emails a summary. Transcription is now near-commodity, and the agent can produce the recap, file it, and update the deal record in one motion instead of dropping a PDF nobody opens.
• List-building tools. The "build me a list of 500 accounts matching this profile" platform. That is a query, and an agent against your data plus public sources answers queries.
• Always-on BI dashboards. Not all of them. But the dashboards built once, watched for a week, then ignored are pure carrying cost. If a question gets asked monthly, ask the agent monthly. Keep the dashboards people actually open.
• Single-feature AI bolt-ons. The fastest-growing category of waste. These wrap a model you already pay for and charge you again for a thin prompt. An email-rewriter, a call-scorer, a subject-line generator. If it is a prompt with a logo, you already own the model underneath.

Tally honestly and the number lands around a third of the stack. That is the 15.

What you keep, and why

Consolidation has a failure mode: cutting the things that actually hold the system together. A few categories earn their keep, and an agent makes them more valuable, not less.

• The system of record. Your CRM is the spine. The agent reads from it and writes to it. Killing it to "let AI handle it" is how you lose your pipeline history. Keep it, and make it the source of truth the agent serves.
• The data layer. The warehouse or whatever holds your clean, joined customer data. This is what makes the agent good instead of generic. An agent with no proprietary data is just a stranger with opinions.
• Deliverability and infrastructure. Domains, inbox warming, sending reputation, the plumbing that gets a message delivered. An agent can write a thousand emails. It cannot will them past a spam filter. This is real engineering and you do not want to own it yourself.
• Anything that owns proprietary data. If a tool holds intent signals, usage telemetry, or a dataset you genuinely cannot reproduce, it is not a wrapper. It is an asset. Keep it.

The pattern: keep what owns data or infrastructure. Cut what owns a workflow the agent can now run.

The method, and its costs

The discipline is one question per tool, asked in order. Map the tool to the job. Ask if the agent already does that job or could with the right access. Then watch for the two things that hide the real bill.

The first is the integration tax. A tool's price is the smaller number. The larger one is the engineering time to wire it in, the breakage when an API changes, and the context-switching cost on every rep who has to remember which of six tools does what. Cutting a tool you barely use still returns that hidden tax.

The second is shadow renewals. The annual contracts that auto-renew on a card nobody checks, for tools whose champion left two quarters ago. Sort your billing by last login. The dead weight surfaces fast, and clearing it can put five figures a year back without touching anything anyone uses.

Be honest that consolidation is not free. There is migration work. There is a transition window where the old tool and the new motion run side by side and you are paying for both. There is the rep who genuinely loves the thing you are cutting and will need a reason, not a memo. Budget for that. A messy consolidation that breaks the rep workflow costs more than the seats it saved.

The goal is not the fewest tools. It is the fewest tools that each do a job the agent cannot, on data the agent needs.

The takeaway

Run the audit this quarter, not next. Pull the billing export, map every tool to its job, and flag the wrappers: standalone enrichment, sequencing add-ons, redundant schedulers, recap bots, list-builders, dashboards nobody opens, and single-feature AI bolt-ons. Keep the system of record, the data layer, the deliverability stack, and anything holding data you cannot rebuild.

You will likely find your 15. The reclaimed budget matters, but the bigger win is a stack a new rep can learn in a day and an agent can actually operate. A smaller surface is a faster one.