· 6 min read
The Sequential Pipeline pattern processes work through a fixed sequence of specialized stages. Each agent in the chain performs a single well-defined function, transforming the work product before passing it to the next stage. The pipeline enforces a strict order of operations: stage N must complete before stage N+1 begins.
This architecture provides two critical guarantees. First, every agent receives fully processed input from all upstream stages, so no agent ever works with incomplete or preliminary data. Second, each stage acts as a quality gate, catching errors before they propagate further into the pipeline. A validation failure at stage two is far cheaper to fix than a corruption discovered after stage five.
The Sequential Pipeline is the right choice when the task has a clear beginning-to-end workflow where each step adds a distinct type of value. The key question is: does step N genuinely need the output of step N-1? If yes, sequential is the correct pattern. If steps are independent, Parallel Workers would be more efficient.
Report generation follows a strict dependency chain that mirrors the way experienced analysts work. You start by gathering data. You analyze that data to find patterns. You write narrative that explains those patterns. You create visualizations that support the narrative. You review everything for accuracy and coherence. Each step requires the output of the previous one.
Attempting to shortcut this sequence produces poor reports. Writing narrative before analysis is complete leads to conclusions that the data does not support. Creating charts before the narrative is finalized produces visualizations that do not align with the story being told. Reviewing a report where the data, analysis, narrative, and visuals were created independently results in an incoherent document.
The Sequential Pipeline enforces the discipline that good report generation requires. Each agent is a specialist: the data agent knows how to query and clean data, the analysis agent knows how to find patterns, the writer knows how to explain findings, the visualization agent knows how to make data comprehensible at a glance, and the reviewer knows how to ensure everything holds together. Chaining them ensures every specialist works with the best possible input.
Data Collection Agent — Gathers, cleans, and structures the raw data needed for the report. This agent connects to specified data sources, executes queries, handles missing values and outliers, normalizes formats, and produces a clean dataset with a data dictionary describing each field. It flags any data quality issues encountered and documents the decisions made during cleaning (for example, how missing values were handled and why).
Statistical Analysis Agent — Receives the clean dataset and performs the analytical work. This agent calculates descriptive statistics, identifies trends and patterns, runs comparisons across segments, tests hypotheses, and surfaces anomalies. It produces an analysis memo that presents each finding with supporting metrics, confidence levels, and caveats. Findings are ranked by significance and business relevance.
Narrative Writer Agent — Takes the analysis memo and crafts the report's written content. This agent writes the executive summary, translates statistical findings into plain-language insights, provides context for why findings matter, draws connections between related findings, and structures the narrative with a logical flow from key takeaways to detailed sections. It maintains the appropriate tone for the target audience (board-level, management, or technical).
Visualization Agent — Receives the narrative draft and the underlying dataset to create supporting visual elements. This agent determines which findings benefit from visual representation, selects the appropriate chart type for each (trend line for time series, bar chart for comparisons, scatter plot for correlations), specifies the exact data to plot, and provides detailed chart specifications including axis labels, legends, annotations, and color encoding. It integrates chart placements into the narrative at the points where they add the most clarity.
Review and Quality Agent — Performs the final quality pass on the complete report. This agent cross-references every claim in the narrative against the analysis memo to verify accuracy, checks that visualizations correctly represent the underlying data, ensures the executive summary accurately reflects the detailed findings, verifies internal consistency (numbers cited in different sections match), and produces a final reviewed version with a quality certification noting any issues found and resolved.
Step 1 — Data gathering and preparation. The Data Collection Agent receives the report specification: "Q1 2026 sales performance report, covering all regions, comparing against Q4 2025 and Q1 2025." It queries the sales database, pulling 47,000 transaction records. It cleans the dataset by removing 312 duplicate entries, filling 89 missing region codes using the customer master table, converting all currencies to USD, and excluding 14 test transactions. It produces the clean dataset with a data dictionary and a cleaning log.
Step 2 — Pattern identification. The Statistical Analysis Agent examines the clean dataset and produces twelve findings. Key discoveries: overall revenue grew 12% quarter-over-quarter and 28% year-over-year; the APAC region grew 34% YoY (fastest); average deal size increased 8% while deal count increased 18%, indicating both expansion and new acquisition; enterprise segment showed 15% higher close rates than mid-market; and three product categories showed declining attach rates despite overall growth. Each finding includes the exact figures, confidence intervals, and comparison benchmarks.
Step 3 — Narrative construction. The Narrative Writer Agent structures the report around the five most significant findings, leading with the overall growth story before drilling into regional and segment details. The executive summary highlights the three most actionable insights in two paragraphs. The detailed sections provide context: the APAC growth story is connected to the regional team expansion in Q3 2025, and the declining attach rates are analyzed for potential causes. The narrative explicitly calls out what leadership should do differently based on the findings.
Step 4 — Visual design. The Visualization Agent creates eight charts integrated into the narrative. A revenue trend line showing four quarters of data with the YoY comparison overlay. A regional performance bar chart comparing all five regions. A deal size distribution histogram showing the shift toward larger deals. A scatter plot of close rate versus deal size by segment. A product category heatmap showing attach rate trends. Each chart specification includes the exact data series, axis ranges, annotations for key callouts, and placement instructions within the document.
Step 5 — Accuracy verification. The Review and Quality Agent performs forty-seven verification checks. It confirms that the "12% QoQ growth" figure in the executive summary matches the calculation in the analysis memo ($14.2M vs $12.7M = 11.8%, rounded to 12%). It catches an inconsistency where the narrative states "three regions grew" but the analysis shows four regions with positive growth (one was near-zero and ambiguously described). It verifies that all chart data points correspond to actual values in the dataset. It produces the corrected final report with a quality log noting four corrections made.
Step 6 — Report package delivery. The pipeline outputs the complete report: a 12-page document with executive summary, five analytical sections, eight integrated visualizations, an appendix with methodology notes and data definitions, and the quality certification. Supporting artifacts include the clean dataset, the analysis memo, and the chart specification files for reproduction.
The final report package would contain the following:
Executive Summary — Two paragraphs covering the headline growth story (28% YoY revenue increase driven by APAC expansion and deal size growth), the primary concern (declining product attach rates in three categories suggest cross-sell motion needs attention), and the recommended action (invest in APAC capacity, investigate attach rate decline root causes, replicate enterprise segment's high close rate practices in mid-market).
Revenue Overview Section — Quarterly revenue trend with four-quarter history, YoY and QoQ growth rates, revenue composition by segment and region, and a waterfall chart showing the components of growth (new customers, expansion, churn).
Regional Performance Section — Five-region comparison with individual narratives for each, highlighting APAC's breakout performance, EMEA's steady growth, and LATAM's underperformance relative to investment.
Deal Dynamics Section — Analysis of average deal size trends, deal count trends, close rate by segment, and sales cycle length changes. Scatter plot showing the relationship between deal size and close rate with segment color coding.
Product Attach Rate Section — Heatmap of attach rates across twelve product categories over four quarters, with three declining categories highlighted and root cause hypotheses presented.
Methodology Appendix — Data sources, extraction dates, cleaning decisions, statistical methods used, and known limitations. Quality certification with verification checklist showing all forty-seven checks passed.