"Pattern-oriented modelling in population ecology" by Grimm et al. (1996), with another useful 'frame' for ways to conduct ABM (in this case anchoring via spatial patterns in the data at particular scales). I (Stuart) just noticed they got a Science paper out of it later in 2005!
Grimm's ODD protocol for describing ABMs was also mentioned, which has been 'adopted' by JASSS as a suggested standard for writing up models in papers (and was 'brought into' computational social science via Polhill et al.).
Evolution of Cooperation by Robert Axelrod. Leads to a very distinct strand in agent-based modelling where very simple agents play well-defined games such as the Prisoner's Dilemma against each other and big (possibly ambitious) conclusions are drawn about human evolution. Notable names in this tradition include Martin Nowak and Karl Sigmund.
Scott Moss and Bruce Edmonds are known for a challenging view in the agent-based modelling area that says there's no reason to expect simple theories to work in the social domain and so we should be building models that include all the factors that "stakeholders" are interested in. This is known as the KIDS approach: "Keep It Descriptive Stupid." This tradition has led to some interesting papers but of necessity the papers tend to be very descriptive and so it's hard to generalize from one model to another.
Economic models: lots of room for agent-based models of markets, economies, etc. Leigh Tesfatsion and Alan Kirman are two people to look at, plus the Santa Fe work on artificial stock markets (e.g., LeBaron (2001)). Look at Tesfatsion's "Agent-Based Computational Economics" website for many useful references.
The work of Benenson and others from Tel Aviv on putting agent-based modelling into a geographical context.
Seminal paper by Oreskes et al. on "Verification, Validation and Confirmation of Numerical Models in the Earth Sciences". Good defence of why ABM models shouldn't be held to higher standards than mathematical models in the never-ending task of verifying and validating models against reality. Relates to a classic comment by Box: "All models are wrong but some are useful".
Brenner and Werker: A Taxonomy of Inference in Simulation Models. One of the few attempts to try to characterise what different styles of model (and different styles of experimentation with them) imply about the inference process to learn about some real-world system (basically validation and some epistemology).
Van Dyke Paranak (get ref. from Jason H.). I (Stuart) think it's this one, comparing ABM with equation-based modelling. (If not, that's a useful 'paradigm comparison' paper.)
Authors such as Brian Skyrms doing models of games on networks. Switching between game-theoretic and agent-based models in this area. Geard, Bryden, et al. wrote a nice paper on a model of homophily in dynamic social networks.
Charles McCall from Argonne Laboratories: military agent-based models and health models. Bruce Lee does similar work on ABMs for health.
George Mason University (e.g., Claudio Cioffi-Revilla) is known for brave/ambitious models of very large social systems with many agents, e.g., the rise and fall of empires in Asia, models of the Afghan opium trade, etc.
John Sterman's Business Dynamics (system dynamics textbook) is a good read for agent-based modellers to realize that operations researchers have covered very similar ground. Lots of convergent evolution between OR, AI, and biology in getting to some of the same ideas about ABMs.
Seminal ideas from the ALife and neuroscience and robotics communities: Tom Ray's "Tierra", Larry Yeager's "Polyworld", Craig Reynolds's "Boids", and Valentino Braitenberg's book "Vehicles". All inspired lots of follow-up work.
Conjunction between the network theory world and the agent-based modelling world when ABM researchers realized they could put their agents on networks as a way of capturing particular patterns of social interaction.
RAND Corporation (Lempert et al.): report on the use of scenario generators (which could be ABMs) in policy-making. "Shaping the Next 100 Years". Argument that you can't build models that are predictively accurate, but instead you should build models (scenario generators that could be a suite of models) that generate as wide a range of plausible futures as possible (without 'picking horses' by assigning probabilities to them), and choose policy that is as robust to *all* of these as possible.
Jonathan Rosenhead: "Rational Analysis for a Problematic World". OR book, but OR has longer experience of pragmatically dealing with real-world complexity.
Lynn Padgett as the representative of an argument that richer models of cognition and agency are needed in ABMs, such as the Belief / Desire / Intention models associated with classical AI.
The "SugarScape" model by Epstein and Axtell. Very interesting book "The Evolution of Artificial Societies" that tries to cover trade, combat, resource collection, etc., and has inspired a lot of similar models.
Josh Epstein's JASSS paper "16 things you can do with a model apart from prediction." ("Why Model"). This bears on the problem of what exactly we think we're doing with these models if prediction of human systems is impossible.