Vijayanta Gupta

Investor note

When the Code Stops Being the Moat, Diligence Has to Find Where It Went

.6 min read

The consensus reading of Bain vibecoding replicas of acquisition targets is that it is a parlour trick. The Financial Times reported on 22 June 2026 that Bain has built hundreds of rough prototypes of target software during private equity diligence, using generative coding tools including Claude Code to recreate core features in days, a practice it started with a dedicated engineering team in 2023 and has since pushed out to ordinary consulting teams (Financial Times via TradingView, 2026; Private Equity Wire, 2026). The loudest objection is that the clones are toys: buggy, insecure, and no substitute for a mature product. That objection litigates the wrong axis. The replica is not a product. It is a diligence instrument, and its job is to locate where the defensibility actually sits once the cost of building the artifact falls toward zero.

That reframing is the underwriting point, and most buyers are still on the wrong side of it.

What the market is mispricing

The SaaS-era underwriting assumption was that a working product, shipped and adopted, was itself the durable asset. Feature depth, code maturity, and release velocity were treated as the moat, because reproducing them took years and capital. When a credible clone of the core workflow takes a weekend, that assumption breaks. The product still has value, but the value has migrated off the artifact and onto position: proprietary data, distribution, switching costs, regulatory surface, and workflow lock-in.

When a working clone takes a weekend, the code was never the moat.

The mispricing is specific. An investor who underwrites the artifact pays for replicable surface area and books it as defensibility. An investor who underwrites position pays only for what a competitor with the same model and a vibecoding team cannot reproduce in a quarter. Those are different prices for the same asset, and the gap between them is the error Bain’s instrument exposes. One investor told the FT that a Bain-built recreation of an analytics platform contributed to their firm dropping out of the bidding, and another summarised the new posture as “if it’s in the question box, we’re not going to touch it” (Startup Fortune, 2026).

The company-level mechanic

This is not a mood about AI. It is already in the prices and the deal flow. Public markets have repriced the incumbents the disruption threatens most: Salesforce is down roughly 37 percent in 2026 and ServiceNow roughly 36 percent year to date, the latter having shed close to half its value since late 2025, on fears that AI agents replace rather than extend the platform (Yahoo Finance, 2026; TIKR, 2026). Both still grow revenue at double digits, which is the point: the market is repricing defensibility, not current performance. On the private side, the value of PE-led technology, telecom, and media deals fell 69 percent in the first quarter of 2026 against the prior quarter, per KPMG data cited by the FT (Crypto Briefing, 2026). Part of that freeze is buyers no longer able to underwrite software moats they can now cheaply test, and declining to bid rather than misprice.

The mechanic that matters at the company level is the replication test. If the core workflow can be cloned to a working state in days, the code is a feature, not the franchise, and the multiple should rest on the data, distribution, and lock-in that survive the clone. If it cannot, the difficulty itself is evidence of a real moat, and the asset deserves the premium the market is currently stripping from the category wholesale.

Upside and risk

The risk to this view is overcorrection. A vibecoded replica recreates visible features, not the operational reality underneath: data networks built over a decade, compliance accreditations, integration depth, and the switching cost of a system embedded in a customer’s close or its supply chain. Read carelessly, the instrument flags a moat as absent when it is merely invisible to a weekend prototype. The replica answers “is the artifact replicable,” not “is the position replicable,” and a diligence team that conflates the two will walk away from durable assets at exactly the moment they are mispriced cheap. The disciplined read cuts both ways: it finds false moats in incumbents and real moats the panic has underpriced.

The upside for a PE-backed platform inside a hold is concrete. The same instrument that prices an acquisition also audits the existing portfolio. Run the replication test against each product line, and the output is a map of which assets are defended by position and which are coasting on a code base a competitor can now clone. That map drives the harder capital-allocation calls a sponsor clock and debt schedule already force: which assets to reprice and defend on data and distribution, which to harvest before the clone arrives, and which to sell rather than transform while the moat still reads as intact to a less sophisticated buyer.

Watch this, not that

Watch replication time, not feature parity. The metric that should move an underwriting decision is how long it takes a competent team with current tools to clone the target’s core workflow to a credible state. Under roughly a quarter, treat the code as undefended and move the entire valuation case onto position, or pass. Over a year, the difficulty is itself the asset.

Three diligence questions follow, each tied to a decision. First: what specifically survives a weekend clone, named as data, distribution, switching cost, or regulatory surface, with the dollar value of each. If nothing survives, the bid comes off the table. Second: is the recurring revenue defended by lock-in or merely by inertia that an AI-native entrant can dissolve, which sets whether the renewal book underwrites the multiple. Third: where does this asset sit on the build-cost curve in eighteen months, not today, because the instrument that priced it can also model where its defensibility goes next. The move each answer triggers is the same discipline: pay for position, never for the artifact, and let the firms still underwriting the code overpay into the freeze.

Sources

Back to writing