An AI Just Filed a Complete 1065 Tax Return on Its Own. What Happens to Accounting Firms Next?

The accounting profession has been told, repeatedly, that AI is a copilot — a tool that assists but never supplants. Basis AI's February 2026 demonstration destroyed that comfortable narrative. The company showed an AI agent completing a Form 1065 partnership return from end to end, without human intervention, on the way to closing a $100 million Series B at a $1.15 billion valuation. The 1065 wasn't chosen arbitrarily. It is one of the most technically demanding filings in the U.S. tax code — requiring individualized K-1 schedules for every partner, custom profit-sharing allocations, and reconciliation across multiple entity types. If an agent can handle that autonomously, the question is no longer whether AI will eliminate compliance-grade accounting work. It's which firms have structured themselves to survive the revenue crater that follows.

What Basis Actually Demonstrated — and Why a 1065 Is the Hardest Possible Starting Point

Choosing the 1065 as the proof-of-concept filing was a deliberate provocation. A straightforward 1040 would have been dismissed as parlor-trick automation. The partnership return is structurally complex: each partner's K-1 must reflect their specific economic arrangement, the return must account for guaranteed payments, basis calculations, separately stated items, and often multi-state apportionment. A typical engagement runs 10–15 hours of professional time for a mid-complexity partnership.

Basis's agents operate in what the company calls "long-horizon" mode — running continuously in the background, coordinating subtasks, and delivering a completed return for human review rather than prompting the human at each step. That's architecturally different from AI-assisted tax software where the CPA still drives every decision. Vinod Khosla's Khosla Ventures, which backed the round, projects 20–50% efficiency gains across practices. With Basis already deployed at roughly 30% of the top 25 U.S. accounting firms, this is not a speculative future state. The technology is inside the walls.

The Liability Vacuum: When an AI Files the Return, Who Signs It?

The answer, under current U.S. professional liability doctrine, is unambiguous and uncomfortable: the CPA whose PTIN appears on the return carries full malpractice exposure, regardless of what generated the underlying numbers. A Morgan Lewis analysis from March 2026 — prompted by Judge Jed Rakoff's February 2026 ruling in United States v. Heppner — confirms that courts are applying traditional professional-duty standards to AI-assisted workflows without modification. The AI vendor is not a licensed professional. The signing CPA is.

This creates a risk asymmetry with no current resolution. Firms absorb the liability; the AI platform collects the SaaS fee. Whether a firm could theoretically pass liability upstream to the vendor depends on whether the system was bespoke or off-the-shelf — and even bespoke implementations face a steep burden of proof. More practically, no engagement letter written before 2025 contemplated autonomous agent involvement. Firms deploying Basis-style platforms on existing clients are likely operating outside their current professional liability coverage without knowing it.

The governance gap is equally alarming. A 2025 KPMG survey found that 46% of U.S. workers admitted uploading sensitive company information to public AI platforms. For CPA firms handling partnership financials, that behavior isn't just a data security failure — it's potential client privilege waiver under the Heppner standard. Firms that haven't built enterprise-grade AI deployment policies aren't just behind on strategy. They're accruing unquantified legal exposure with every client file that touches a public model.

The Billing Model Problem: Hourly Rates Don't Survive Automation

The pricing reckoning is already underway, and autonomous agents will accelerate it past the point of no return. Only 3% of accounting firms now bill purely by the hour for tax preparation — a collapse from the historical norm driven by firms that discovered AI-compressed timelines made hourly billing actively harmful to revenue. The shift to fixed-fee and value-based models has been underway, but it hasn't fully resolved the underlying problem: if the cost of producing a completed 1065 drops from 12 hours of CPA time to 20 minutes of agent time plus 2 hours of review, what justifies the fixed fee you set last year?

Bloomberg Tax's reporting on Big Four AI investment illustrates the stakes: RSM committed $1 billion over three years in technology spending. The margin they're capturing by automating compliance work must be recycled into pricing strategies that don't cannibalize partner revenue. The firms getting this right are engineering a specific outcome — using AI efficiency to expand advisory capacity rather than reduce fees. The CPA Trendlines 2026 Outlook quantifies the reward: tech-optimized firms achieve $250,000–$350,000 in revenue per employee annually, against $150,000–$200,000 for firms on traditional models. That is a structural profitability gap that compounds year-over-year.

98% of Firms Already Use AI Daily — So Why Aren't They Ready for This?

Karbon's 2026 State of AI in Accounting report, surveying nearly 600 professionals across six continents, found that 98% of accounting firms now use AI, generating an average of 60 minutes of productivity savings per employee per day. By any measure, the profession has embraced AI tooling at scale. Yet the same report reveals that only 21% have documented AI policies or strategies, and fewer than 50% invest in formal training. Firms have industrialized AI adoption without industrializing AI governance.

This bifurcation matters enormously when autonomous agents enter the picture. Point-tool AI — a ChatGPT prompt here, a document summarizer there — carries contained risk. An agent that autonomously completes a 1065 and routes it for a single sign-off review requires a fundamentally different governance architecture: defined review checkpoints, explainability documentation, audit trails, client disclosure protocols, and liability-aware engagement terms. Firms deploying Basis without those structures aren't being innovative. They're taking on concentrated professional risk in exchange for efficiency gains they haven't yet figured out how to monetize.

Thomson Reuters' research shows that firms with documented AI strategies are 3–4 times more likely to see measurable revenue growth and efficiency gains. The 79% of firms operating without those strategies are not simply behind — they are running a system that becomes structurally disadvantaged faster as agents raise the capability ceiling.

The New Division of Labor: What 'Professional Judgment' Means When Agents Do the Work

AICPA CEO Mark Koziel's oft-quoted reassurance — that AI will not disrupt the accountant as long as the profession keeps skills current — is technically accurate but strategically incomplete. The skills that must be kept current are not the skills that built the profession. Preparing a 1065 from source documents is a mechanical skill an agent now executes faster and with comparable accuracy. The professional judgment that justifies licensure is the layer above that: identifying when an autonomous output reflects a planning opportunity the agent didn't flag, catching structural errors that require entity-level context the agent lacks, and bearing the legal and ethical accountability for the final signed return.

PwC's Dan Priest has argued that AI will reverse the accounting profession's drift toward narrow specialization, enabling CPAs to function as broader advisors. That's the correct strategic frame — but it requires firms to build the incentive structures and training pipelines that support generalist advisory capacity. The downstream talent risk is real: junior accountants who never prepare returns manually may lack the foundational understanding needed to catch agent errors. As Accounting Today has noted, experts warn that entry-level staff losing exposure to foundational work will struggle to evaluate AI outputs, requiring stronger mentorship architecture. The profession cannot simultaneously automate entry-level work and produce the next generation of professionals who can supervise that automation competently.

Which Firm Archetypes Survive the Autonomous-Agent Era — and Which Don't

The outcome is not uniform across firm types, and the divergence will be visible within 24 months. Large regionals and Big Four-adjacent firms that have already invested in proprietary AI infrastructure and rebuilt pricing around advisory retainers are structurally positioned to capture margin compression as a competitive advantage. They automate compliance cheaply, redeploy capacity into higher-fee strategic work, and use the efficiency differential to undercut mid-market firms on compliance pricing while dominating on advisory scope.

The firms in genuine existential danger are the compliance-heavy mid-tier practices — roughly 50–200 employees — that generate the majority of their revenue from exactly the filing categories agents now handle autonomously. These firms face the worst possible combination: they lack the scale to invest in proprietary AI infrastructure, they've built client relationships on compliance delivery rather than strategic advisory, and their billing models are calibrated to labor costs that are about to crater. CPA Trendlines projects the competitive gap becomes "insurmountable" by 2028 for firms without clear AI strategies — a timeline that aligns precisely with the projected maturation of autonomous agent deployment at scale.

Small boutique practices with deep advisory specialization in specific industries or entity types — private equity fund accounting, real estate partnership structures, cross-border tax planning — face a different trajectory. Their moat is domain expertise and relationship capital that agents cannot yet replicate. They are at risk not from automation but from the larger firms using automation-funded capacity expansion to compete down-market into their niches.

The 1065 demo was a technical milestone. The firm-level reckoning it forces is a strategic one. Who carries the liability, how you price a 20-minute agent output, and what you tell your junior staff they're training for — those questions don't have comfortable answers yet. The firms that survive will be the ones that stopped waiting for comfortable answers in 2026.