In a workshop last week, someone asked me the question everyone’s been circling: If AGI becomes a thing, will we still need CPQ systems? It landed with the kind of silence you only get when a room knows the question matters.

I’ve worked with CPQ since 2000, mostly with Tacton, across demos, delivery, governance, and scaling. I’ve seen the pattern repeat: CPQ nails validation and structure, then stalls at the moment that actually moves deals forward. The salesperson still has to answer the human question the buyer is really asking: Why this configuration for my situation?

Large language models can talk about “why.” They can reason about scenarios, surface trade-offs, and make the complex relatable. That changes the shape of the risk. The threat is not that AI replaces CPQ. The threat is that AI exposes where CPQ currently stops.

The Wrong Question About CPQ Obsolescence

The binary frame - AI replaces CPQ or it doesn’t - hides the real design problem. Traditional CPQ excels at correctness, governance, and repeatability. It is deterministic. Same input, same output. It guards what is allowed and calculates what it will cost. That matters deeply in complex manufacturing, where buildability is the line between margin and mayhem.

But the sales moment rarely starts with a part number. It starts with a scenario: tight urban streets, intermittent off-road access, four crew, and a five-year TCO conversation. LLMs are good at this kind of reasoning. They can unpack context, compare options, and express trade-offs in plain language. They do not replace the hard boundary of what is valid, but they compress the time it takes to reach an informed, defensible choice.

If your quoting path cannot explain itself while it moves, the field will route around it.

Look at the market signals. Salesforce has put its legacy CPQ into end-of-sale for new customers, steering buyers to a new Revenue Cloud stack at higher price points. As one industry post put it, “Salesforce CPQ has officially entered ‘end-of-sale’ status for new customers.” (The CPQ Crisis: Why Your Company Has Just 5 Years to Get Off Salesforce CPQ, LinkedIn). According to Salesforce’s own pricing page, Revenue Cloud Advanced lists at $200 per user per month (Salesforce Pricing Page). That is not just a SKU change. It’s a forced migration moment that makes every CPQ leader ask: if we are moving anyway, what are we moving toward?

If your answer is simply “the next CPQ,” you will miss the structural shift. The capability boundary is moving upstream, from validation to reasoning.

Why This Moment Is Different

For twenty years, teams have accepted a trade-off: speed vs trust. Excel is fast but brittle. CPQ is trustworthy but often slow to adapt and slow to explain. LLMs scramble that equation. They bring a new interface to the same hard problems: sense-making, comparison, and narrative clarity, all in the buyer’s language.

Two constraints keep us honest. First, LLMs are probabilistic. They can generalize, rank options, and draft explanations, but they can also be confidently wrong. Second, enterprise sales needs guarantees. You do not build MRI machines, industrial trucks, or process skids with probabilistic rules.

So the winning shape is emerging: use explicit, testable logic for validity and cost, and let language models handle intent capture, scenario reasoning, and explanation. The system remains the guardrail. The new layer is the guide.

Rules guarantee correctness. Language models compress time.

Why it matters now, not later:

  • Vendor-driven change has shortened the decision window. Forced migration is not a theoretical risk; it’s on renewal calendars right now.
  • Cost pressure is real. $200 per user per month for advanced quoting capability concentrates minds on actual daily use and measurable impact.
  • Buyer behavior has shifted. Teams expect a system that both reasons and shows its work. If the sanctioned path lags behind spreadsheets and ChatGPT, shadow quoting becomes standard practice.

A Hybrid Reasoning Architecture For CPQ

Here is the architecture that resolves the old trade-off without betting the company on hype. Think of it as a reasoning layer on top of explicit logic:

1) Explicit product logic as foundation. Keep rules, constraints, and price drivers where they belong: in a system that is deterministic, testable, and owned. Equality constraints, scenario modules, and structured data win because they scale and can be verified. This is where BOM, buildability, and pocket price earn their keep.

2) A conversational reasoning layer on top. Use LLMs to capture intent, compare options, and narrate trade-offs. This is where questions like “Why pick this gearbox for urban duty cycles?” actually get answered, with references back to the explicit logic. The LLM proposes; the rules dispose.

3) Explanation by design. Every recommendation must be paired with a verifiable justification: constraint reasons, cost impacts over time, and scenario suitability. If the system cannot show why a choice is valid and preferable, it is not production-ready.

4) Guardrails and test suites. Treat the LLM like a junior consultant: powerful with supervision. Route its proposals through the constraint engine. Maintain a living test suite of your top quoting paths. When the model, price list, or prompts change, rerun the suite before the field feels it.

5) Progressive data strategy. Start small but structured. Write the product in sales English that is specific enough to be reasoned on: scenarios, pros and cons, do-use and don’t-use. Keep the context tight. As your usage grows, expand the corpus and refine prompts. Progress beats perfection.

Practically, this means your roadmap shifts from a single monolithic rollout to a layered capability:

  • Near term: Put a reasoning interface in front of 2-3 high-volume quoting paths. Keep the current CPQ or ERP logic as the arbiter.
  • Next: Instrument explanations. Expose constraint reasons, TCO deltas, and price movements on-screen. Make the system teach as it quotes.
  • Then: Build governance around change velocity. Move ownership of text and scenarios to product and sales ops. Keep rules and tests with your CPQ team. Shorten the draft-to-live cycle from months to days.

Who wins in this model? The teams that accept that narrative is part of configuration. They make the system both reason and explain in the buyer’s terms, with rules quietly ensuring nothing silly gets through. They measure adoption weekly because adoption is the only metric that matters.

Who drifts? The programs that burn a year on tool migration while leaving the top five objections unanswered in the product experience. The teams betting on a pure LLM without guardrails will find out the hard way that guesswork at scale does not sell capital equipment. The teams doing nothing will wake up to shadow workflows where reps pair spreadsheets with chat assistants because the official path is still slower.

This is not about choosing a vendor. It is about choosing an architecture that accepts what each technology is good at. Determinism handles validity. Probabilistic reasoning handles language, comparison, and learning. Put them in the right order and you get speed with trust.

I do not believe AGI will erase the need for CPQ in complex B2B. I do believe the definition of CPQ will change. The hard boundary moves upstream to include scenario reasoning and explanation as first-class capabilities, not add-ons. That is the moment we are in.

The quiet risk is assuming this is a futures topic. It is already here in field behavior. Buyers expect to understand trade-offs instantly. Reps expect the system to help them think, not just click. Finance expects a clear line from configuration to cost and margin. If one layer fails, the deal decays in days, not weeks.

So ask a different question than the one that started this post. Not will AI make CPQ obsolete, but where, exactly, does your current system stop helping a buyer decide? And if an assistant can now fill that gap in minutes while your rules keep it safe, what are you waiting for?

When your system can both reason and show its work, what part of your quote still needs a spreadsheet to feel believable?