When an agency asked for a deck, we sent them the codebase instead

May 7, 2026

When an agency asked for a deck, we sent them the codebase instead

A partner agency emailed last month asking for a slide deck. Something to walk their team through how our automation work translates to client outcomes. Standard ask. Most agencies have a polished 15-slide answer ready for this exact moment. We did not have a deck. We sent a Loom of the Atlas dashboard, a link to this blog, and a paragraph: here is the work, here is how it ships, here is the activity feed for the last 30 days. The codebase is the deck. This piece is for operators of services businesses who suspect the deck is the bottleneck and want a working alternative.

The follow-up came the next day with the question that mattered. Were we an agency, a product company, or something in between? The honest answer is that Arthea is an AI-native operating system that sells outcomes. The agency framing is a legacy of how the industry used to be priced. The product framing is what the work has actually become. "Something in between" is what most of the market is still saying because they have not finished the transition.

When the codebase is the deck: a positioning shift for AI-native services

The codebase is the deck because the artifact does the talking. A working dashboard, a live activity feed, a draft queue arriving in the prospect's voice answers more questions in five minutes than fifteen slides can in an hour. The shift only works if the artifact is real. If the dashboard is a prototype, the deck is still load-bearing and you have not actually moved.

Most agencies cannot make this move because the work behind the deck is bespoke. Every engagement is rebuilt from a blank page. The deck has to do the explaining because the work itself is unrepeatable. When the work is reproducible, the artifact replaces the explanation. That is the whole shift in one sentence.

Why most agencies still need decks

A deck is what you write when the work cannot be shown. It abstracts the engagement into a series of slides because the engagement itself is bespoke and unobservable from the outside. The slide format hides the lack of repeatability. Fifteen slides on "our process" is a substitute for an artifact that demonstrates the process is real.

There is nothing wrong with this when the work genuinely is bespoke. A high-end consultancy with a senior partner running a six-month engagement is not going to ship a self-serve demo. But that is a narrow band of services. Most agency work has more reproducibility than the deck format admits, and the deck is doing the wrong job.

Why this matters for AI-native services

AI-native services have a structural advantage. The orchestration layer, the brief schemas, the agent stacks, the dashboards: all of it is reproducible by construction. Every prospect can be provisioned a personalised instance in 90 seconds. Every prospect can see the activity feed run live for a week before deciding. The gap between "explain the work" and "show the work" closes.

This is the move that turns an AI agency into a product company without the founder needing to call the shot from the top down. The work itself does the conversion. The deck becomes vestigial.

The framework: artifact-led sales for reproducible work

Provision a working instance. Every prospect gets a personalised Atlas-style workspace, pre-loaded with their briefs, running for a week. The provision is short. The instance is real, not a sandbox. Operators see drafts arrive in their queue with their voice. The conversation starts at "what should the agents focus on for our team" instead of "how do you work."

Publish the activity feed. A 30-day window of what shipped, what broke, what is on next week. The feed is unfiltered. Visible mistakes are part of the credibility. A feed that only shows wins reads like a deck.

Show the dashboard. Live, with real metrics. The dashboard is the surface area where prospects see how the system thinks. They probe it. They ask why a number moved. The conversation is operational, not ceremonial.

Keep one paragraph of context. The Loom and the blog and the activity feed are the artifacts. One paragraph orients the prospect: who you are, what you build, where to look first. Anything longer is decoration.

Skip the deck entirely. If a prospect insists on a deck, that is a useful signal. Either the artifact is not yet ready to do the talking, or the prospect is wired for a different kind of relationship and is not your fit.

Runbook: replacing the sales deck with a working artifact

1. Map every claim your current deck makes about the work. "We move fast", "we have proprietary AI", "our process is rigorous". For each claim, identify the artifact in the live product that proves it. If a claim has no artifact behind it, drop the claim. 2. Build a 90-second provisioning flow for prospect instances. Same templates, same agents, same dashboard, pre-loaded with the prospect's brand inputs. The first time it takes 90 minutes to build. After that, 90 seconds. 3. Publish the activity feed publicly. Last 30 days of what shipped, what broke, what is on next week. No filtering of mistakes. 4. Record one Loom of a prospect-style walkthrough. Six minutes, one take, no editing. Every subsequent prospect gets the same Loom. It does not personalise. It does not need to. 5. Replace the deck-send step in your sales sequence with: Loom + activity feed link + provisioning link + one paragraph. The paragraph is reused. The links are personalised once. 6. Track close rate and cycle length for the next 30 days. Compare against your baseline. The cycle should shorten because the artifact answers questions the prospect would have raised on call two. 7. After 90 days, if a prospect still asks for a deck, audit which questions the artifact failed to answer. Fix the artifact. Do not write the deck.

When this is wrong: the trade-offs

This breaks if the work is genuinely bespoke. A six-month strategy engagement with a senior partner does not provision a working instance. The deck is doing real work there because the engagement itself does not exist outside the conversation. Forcing the codebase-as-deck move on a non-reproducible service produces a bad demo of a thing that was never meant to be demoed.

It also breaks if the dashboard is a prototype. A demo that fakes the activity feed is worse than no demo. Prospects probe live products. If the feed has gaps the artifact is exposed in the first ten minutes and the trust cost is permanent. Ship the deck instead, finish the artifact, then make the move.

The deeper risk is mistaking polish for proof. A beautifully designed artifact that does not actually run is the same thing as a beautifully designed deck. The only thing that matters is whether the work is reproducible and observable. The format is downstream.

What success looks like

When the artifact replaces the deck, three things move. First, close rates go up because the prospect has answered most of their own objections by the time they reach the second call. Second, sales cycles shorten because the conversation skips "how do you work" entirely. Third, pricing moves from billable hours to outcomes per month, and cost structure compresses because labour stops being the constraint.

On the engagements past the 30 to 50K EUR per month band, the artifact is what makes the pricing model legible. Outcomes-per-month is a foreign concept until the prospect has watched the agents run for a week. After that, the conversation is "what is the right outcome target", not "how many hours will this take." That shift is the entire point.

The qualitative band is meaningful. The agency framing starts to feel like an expensive habit instead of a category. The deeper shift is that the work itself becomes the marketing.

FAQ

Does this only work for AI-native businesses? It works for any services business with reproducible work. AI-native is a structural advantage because the orchestration layer is reproducible by construction, but a productised consulting firm with a real artifact behind the engagement can run the same play. The test is whether the work can be shown, not whether AI is involved.

What about prospects who insist on a deck? Send the artifact anyway. If they still insist, that is signal. Either the artifact is not yet ready to do the talking, or the prospect is wired for a different kind of relationship and is not your fit. The job is not to make every prospect close. The job is to close the right ones faster.

How long does the provisioning take to build? The first instance takes 90 minutes because you are inventing the templates. By the tenth, the provisioning runs in 90 seconds because the templates are stable. Most of the cost is upfront and amortises across every prospect after.

Does this hurt premium pricing? The opposite. Showing the work makes outcomes-per-month pricing legible. Premium pricing in services is gated by trust, and an artifact that runs live is a faster trust transfer than any deck. Pricing follows the artifact.

What if the artifact has bugs? Visible mistakes in the activity feed are part of the credibility. A feed that only shows wins reads like a deck. The bar is not zero bugs. The bar is honest reporting and a fast fix loop the prospect can see.

- https://www.arthea.ai/article/this-week-in-arthea - https://www.arthea.ai/article/three-agents-shipped-meaningful-work - https://www.arthea.ai/ai-lab

If you want a 30-minute review of whether your work is reproducible enough to replace the deck, the calendar is here: arthea.ai/book.

Go back to Blog

Download the ressource